source-checked milestone · 12 June 2017 · Vaswani et al.
Decision: FoundationEvidence: Research resultAustralian access: not applicable
The original paper introduces a different architecture for sequence transduction.
Why it is a row
Evaluate attention-based sequence models as an alternative architecture for language-processing work.
What this is not
This is a research foundation, not a hosted product or an autonomous-agent release. Later paper revisions are not new milestones.
Read the decision guides
Primary source: Vaswani et al.
Source support and record
- Record id
- transformer
- Event date
- 2017-06-12
- Source date
- 2017-06-12
- Review status
- source-checked · 2026-09-26
- Era
- foundations
Foundation — source support (close paraphrase): The authors propose the Transformer, an architecture using attention without recurrence or convolution. Location: Version 1 abstract and submission history.
source-checked milestone · 11 June 2020 · OpenAI
General-purpose language models through an API
Decision: AccessEvidence: early accessAustralian access: unresolved
OpenAI opens applications for a general-purpose text-in, text-out API using GPT-3-family weights.
Why it is a row
Consider applying for hosted model access to build a language feature without operating model infrastructure.
What this is not
An application path is not immediate access. Australian eligibility and suitability for client data are unresolved.
Selection: identity and access
OpenAI’s 11 June 2020 announcement introduced API access in a private beta, with a request-for-access path.
Next check
Identify the API service and evaluation account separately from any chat subscription; record the model identifier or selection policy.
Country and account access were not established by an application path.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Opening/API private-beta paragraphs. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- openai-api
- Event date
- 2020-06-11
- Source date
- 2020-06-11
- Review status
- source-checked · 2026-09-26
- Era
- foundations
Access — source support (close paraphrase): Users can request API access to integrate the models into a product; the launch is a private beta. Location: Opening body and private-beta paragraph.
source-checked milestone · 29 June 2021 · GitHub
Copilot brings suggestions into the coding workflow
Decision: PlaceEvidence: early accessAustralian access: unresolved
GitHub launches a limited technical early access of its AI pair programmer.
Why it is a row
Evaluate suggestions made while a developer writes code, with the developer reviewing each proposed change.
What this is not
This is assisted coding, not permission to run a repository unattended. The February 2022 page update is not another launch.
Read the decision guides
Primary source: GitHub
Source support and record
- Record id
- copilot
- Event date
- 2021-06-29
- Source date
- 2021-06-29
- Review status
- source-checked · 2026-09-26
- Era
- foundations
Place — source support (close paraphrase): Copilot uses the code being worked on as context and suggests lines or functions as the developer types. Location: Launch body; original date above byline.
source-checked milestone · 4 Mar 2022 · OpenAI researchers / arXiv
InstructGPT paper submitted
Decision: FoundationEvidence: Research resultAustralian access: not applicable
Training language models to follow instructions with human feedback was first submitted to arXiv on 4 March 2022. Diogo Almeida is a listed coauthor.
Why it is a row
Provides a dated research reference and the supported authorship credit for Diogo Almeida. It is research context, not evidence of Jev accuracy or an inventor title.
What this is not
This is the paper submission date, not a ChatGPT release date. It establishes the research credit; it does not establish a Jev capability, outcome or accuracy result.
Read the decision guides
Primary source: OpenAI researchers / arXiv
Source support and record
- Record id
- instructgpt
- Event date
- 2022-03-04
- Source date
- 2022-03-04
- Review status
- source-checked · 2026-10-07
- Era
- public
Foundation — source support (close paraphrase): The paper describes training language models to follow instructions with human feedback and lists Diogo Almeida among its authors. Location: Title, author list and abstract; submission history v1: 4 March 2022, 07:04:42 UTC.
source-checked milestone · 30 Nov 2022 · OpenAI
ChatGPT opens a conversational research release
Decision: PlaceEvidence: early accessAustralian access: unresolved
The dated announcement offers a conversational product for user feedback.
Why it is a row
Reconsider whether staff can evaluate a conversational AI interface directly, without first building an application.
What this is not
The page does not establish Australian account access or approval to submit client material.
Selection: identity and access
OpenAI’s 30 November 2022 announcement introduced a conversational research release of ChatGPT.
Next check
Record the chat product, plan and intended account. Evaluate a proposed API workflow as a separate configured candidate.
A conversational release is separate from a proposed API integration; current plans and access need current evidence.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Introduction and Iterative deployment. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- chatgpt
- Event date
- 2022-11-30
- Source date
- 2022-11-30
- Review status
- source-checked · 2026-09-26
- Era
- public
Place — source support (close paraphrase): OpenAI introduces ChatGPT as a research release that people can try through its conversational interface. Location: research release introduction and Iterative deployment.
source-checked milestone · 20 Mar 2023 · OpenAI
ChatGPT cross-user history exposure
Decision: TrustEvidence: AnnouncementAustralian access: unresolved
OpenAI’s incident report documents a cross-user data exposure and the subsequent fix.
Why it is a row
Reassess which client information may enter a shared chat service and how affected material would be identified.
What this is not
20 March is the incident day; the report was published on 24 March. This historical event is not evidence of a current leak or that every user was affected.
Primary source: OpenAI
Source support and record
- Record id
- chatgpt-history-incident
- Event date
- 2023-03-20
- Source date
- 2023-03-24
- Review status
- source-checked · 2026-09-26
- Era
- public
Trust — source support (close paraphrase): A bug allowed some users to see another active user’s chat titles and potentially a newly created conversation’s first message. Location: Opening incident description and Technical details.
source-checked milestone · 18 July 2023 · Meta
Llama 2 weights for research and commercial evaluation
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
Meta’s announcement explicitly includes pretrained and conversational model weights.
Why it is a row
Reconsider hosted access versus running supplied weights, subject to the exact Llama 2 licence and use policy.
What this is not
The announcement is not an unrestricted licence grant or a finding that a client workload is permitted. Exact terms still need review.
Weights and permitted use
Llama 2 Community License · Llama 2 materials
The licence grants use, modification and distribution under conditions. Redistribution requires the agreement and attribution notice; the acceptable-use policy is incorporated. Section 1.b.v restricts using Llama materials or outputs to improve other large language models, excluding Llama 2 and its derivatives.
Next check
Check the intended business activity, incorporated acceptable-use policy and redistribution obligations. The additional commercial gate applies above 700 million monthly active users measured as specified at release; this is not an eligibility finding for your firm.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself
Licence passages: §1.a, §1.b.i–v, §2. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Meta
Source support and record
- Record id
- llama
- Event date
- 2023-07-18
- Source date
- 2023-07-18
- Review status
- source-checked · 2026-09-26
- Era
- public
Licence — source support (close paraphrase): Meta makes Llama 2 available for research and commercial use and includes weights and starting code. Location: Body paragraph beginning “We’re now ready”.
source-checked milestone · 28 Aug 2023 · OpenAI
Enterprise data and identity controls
Decision: ControlEvidence: AnnouncementAustralian access: unresolved
ChatGPT Enterprise launches with stated business-data and organisation-management protections.
Why it is a row
Evaluate an organisation-managed chat environment against client confidentiality and staff-access requirements.
What this is not
These product commitments do not establish Australian data residency, contractual suitability or a compliance conclusion.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- chatgpt-enterprise
- Event date
- 2023-08-28
- Source date
- 2023-08-28
- Review status
- source-checked · 2026-09-26
- Era
- public
Control — source support (close paraphrase): OpenAI says Enterprise business data and conversations are not used for training and describes SSO, domain verification and admin controls. Location: Protect your data with enterprise-grade privacy, security, and deployment tools.
source-checked milestone · 11 Dec 2023 · Mistral AI
Mixtral introduces permissively licensed MoE weights
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
The release makes a sparse model family available with a named permissive licence.
Why it is a row
Evaluate a self-operated mixture-of-experts model under Apache 2.0 as another deployment option.
What this is not
The post’s benchmark and inference-efficiency claims are not a verified cost saving for a firm. No Cost secondary type is adopted.
Weights and permitted use
Apache License 2.0 · Mixtral-8x7B-v0.1
Mistral’s release and pinned model card identify Apache 2.0. The licence grants copyright permissions for modification and distribution, with conditions for supplying the licence, marking changed files and preserving applicable notices. It does not grant general trademark rights or a warranty.
Next check
Retain the exact checkpoint and applicable notices. Match the intended modification and delivery to the licence; separate runtime, data and service terms still need checking.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself · Pinned publisher model card
Licence passages: §2–4, §6–7. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Mistral AI
Source support and record
- Record id
- mixtral-8x7b
- Event date
- 2023-12-11
- Source date
- 2023-12-11
- Review status
- source-checked · 2026-09-26
- Era
- public
Licence — source support (close paraphrase): Mistral releases the open-weight Mixtral 8x7B sparse mixture-of-experts model under Apache 2.0. Location: Opening release paragraph.
source-checked milestone · 15 Feb 2024 · Google DeepMind
A limited early access of million-token context
Decision: AccessEvidence: early accessAustralian access: unresolved
Google opens restricted testing of Gemini 1.5 Pro’s enlarged context window.
Why it is a row
Consider a long-document evaluation through the private early access rather than assume the expanded context is generally available.
What this is not
Input capacity is not elapsed autonomous working time, perfect recall or universal account access. The snippet’s Capability classification is not retained.
Selection: identity and access
Google’s 15 February 2024 announcement described limited Gemini 1.5 Pro early testing through AI Studio and Vertex AI.
Next check
Record the exact model, interface and access stage. Do not treat an input-capacity announcement as tested task fitness.
The announcement described private-preview access for a limited group of developers and enterprise customers via AI Studio and Vertex AI. It does not establish current account access or task fitness.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Introducing Gemini 1.5, limited developer/enterprise testing paragraph. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Google DeepMind
Source support and record
- Record id
- gemini-15-preview
- Event date
- 2024-02-15
- Source date
- 2024-02-15
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): A limited group of developers and enterprise customers can try up to one million tokens in AI Studio and Vertex AI private early access. Location: Introducing Gemini 1.5, private early access paragraph.
source-checked milestone · 26 Feb 2024 · Mistral AI
Mistral Large becomes available through Azure
Decision: AccessEvidence: AnnouncementAustralian access: unresolved
Mistral names Azure as its first distribution partner.
Why it is a row
Evaluate the new Azure access path alongside Mistral’s own platform when choosing a provider.
What this is not
The announcement does not settle Australian regions, account entitlements or data-handling terms.
Selection: identity and access
Mistral’s 26 February 2024 announcement offered Mistral Large through la Plateforme and Azure.
Next check
Record which provider route is offered, who contracts and who hosts; ask for account and region evidence for that route.
Two named distribution routes do not identify your contracting party, Australian region, intended account or data-handling terms.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Opening distribution paragraph. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Mistral AI
Source support and record
- Record id
- mistral-azure
- Event date
- 2024-02-26
- Source date
- 2024-02-26
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): Mistral Large is available that day through la Plateforme and Azure. Location: Try Mistral Large and Mistral Small today.
source-checked milestone · 4 Mar 2024 · Anthropic
Claude API becomes generally available
Decision: AccessEvidence: GAAustralian access: unresolved
The Claude 3 announcement changes the API access path as well as launching models.
Why it is a row
Reconsider direct API evaluation now that developers can sign up for the generally available service.
What this is not
Haiku was still forthcoming. A 159-country statement alone does not verify Australian access; model rankings are not the reason for this row.
Selection: identity and access
Anthropic’s 4 March 2024 Claude 3 announcement described Opus and Sonnet in its generally available API, with Haiku forthcoming.
Next check
Check the exact model and route. Do not transfer one model’s announced availability to every member of its family.
Availability was model- and route-specific; the historical announcement does not verify current Australian account access.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Model availability. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- claude-api-ga
- Event date
- 2024-03-04
- Source date
- 2024-03-04
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): Opus and Sonnet are available through the API, now generally available, with developers able to sign up immediately. Location: Model availability.
source-checked milestone · 18 Apr 2024 · Meta
Decision: AccessEvidence: AnnouncementAustralian access: stated
The announcement adds named countries to the assistant rollout.
Why it is a row
Review staff use of AI inside familiar messaging and social apps as the English-language rollout reaches Australia.
What this is not
A rollout is not proof that every account had access that day or that client work is permitted. The page also displays a 3 July update date.
Meta’s rollout page dated 18 April 2024 explicitly names Australia for English-language Meta AI; it also shows a 3 July update.
Next check
Record Australia as named in this historical rollout, then check the current interface, tier and intended account separately.
The page also displays a 3 July update date. A rollout is not proof that every account had access on the announcement date or has access today.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Opening country-rollout paragraph. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Meta
Source support and record
- Record id
- meta-ai-australia
- Event date
- 2024-04-18
- Source date
- 2024-04-18
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): Meta says it is rolling out English-language Meta AI beyond the US and explicitly lists Australia. Location: Opening country-rollout paragraph.
source-checked milestone · 29 May 2024 · Mistral AI
Codestral introduces non-production licence terms
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
Mistral introduces a licence family distinct from its Apache 2.0 releases.
Why it is a row
Check the non-production terms before treating downloadable coding weights as available for business deployment.
What this is not
This records the issuer’s stated boundary, not a legal opinion on any proposed use.
Weights and permitted use
Mistral AI Non-Production License 0.1 · Codestral-22B-v0.1
The pinned model card points to MNPL 0.1. Its permitted testing, research, personal and evaluation activities are limited to a non-production environment. The definition excludes live business operations and end-user use by employees or customers. Outputs remain subject to the use restrictions.
Next check
Do not treat an internal staff workflow as cleared merely because it is private or called a pilot. Identify the exact proposed activity and obtain separate permission where the non-production grant does not cover it.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself · Pinned publisher model card
Licence passages: §3.2–3.3, §4.2, §9 definitions. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Mistral AI
Source support and record
- Record id
- codestral-mnpl
- Event date
- 2024-05-29
- Source date
- 2024-05-29
- Review status
- source-checked · 2026-09-26
- Era
- agents
Licence — source support (close paraphrase): The new MNPL allows non-commercial and research use; Codestral is released under it that day. Location: MNPL introduction and Codestral release paragraphs.
source-checked milestone · 21 June 2024 · Anthropic
Tool-equipped coding with Claude 3.5 Sonnet
Decision: CapabilityEvidence: AnnouncementAustralian access: unresolved
Anthropic describes a coding workflow with tool execution and troubleshooting.
Why it is a row
Evaluate a bounded coding task where the model can edit and execute code, rather than only suggest it.
What this is not
The statement is conditional on tools and instructions. Benchmark gains and unchanged mid-tier pricing do not create a separate Cost decision.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- sonnet-35-agentic
- Event date
- 2024-06-21
- Source date
- 2024-06-21
- Review status
- source-checked · 2026-09-26
- Era
- agents
Capability — source support (close paraphrase): With instructions and appropriate tools, Claude 3.5 Sonnet can independently write, edit and execute code. Location: Frontier intelligence at 2x the speed.
source-checked milestone · 23 July 2024 · Meta
Llama outputs may improve other models
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
The release contains an explicit change to permitted downstream model work.
Why it is a row
Reconsider using Llama-generated material to improve another model under the revised terms.
What this is not
This is narrower than an unrestricted training-data or licence conclusion; other conditions still apply.
Weights and permitted use
Llama 3.1 Community License · Llama 3.1 materials
This licence states a grant for use, modification and distribution with conditions. Making covered materials, derivatives, products or services available brings agreement and Built with Llama obligations. When materials or outputs improve an AI model that is distributed or made available, its name must begin with Llama. Attribution and the incorporated acceptable-use policy also matter.
Next check
Read the actual 3.1 terms for the planned output reuse and delivery. Do not carry Llama 2’s other-model restriction across by assumption; the 3.1 text differs. Check the separately specified 700-million-user commercial gate and account for all applicable conditions.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself
Licence passages: §1.a, §1.b.i–iv, §2. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Meta
Source support and record
- Record id
- llama-31-licence
- Event date
- 2024-07-23
- Source date
- 2024-07-23
- Review status
- source-checked · 2026-09-26
- Era
- agents
Licence — source support (close paraphrase): Meta says it changed the licence to let developers use Llama outputs, including 405B outputs, to improve other models. Location: Introducing Llama 3.1, licence-change paragraph.
source-checked milestone · 24 July 2024 · Mistral AI
Large 2 self-deployment has a commercial licence gate
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
Mistral distinguishes research rights from commercial self-deployment rights.
Why it is a row
Budget for a separate commercial licence before choosing Large 2 for commercial self-deployment.
What this is not
The same page’s general older-model deprecation statement lacks an exact affected inventory and sunset date; it is not attached as Retirement.
Weights and permitted use
Mistral AI Research License 0.1 · Mistral-Large-Instruct-2407
The pinned model card links MRL 0.1. It restricts models, derivatives and outputs to defined research purposes: non-commercial, non-profit research unconnected to business operations. The definition excludes staff’s daily company work and testing intended to generate revenue. The historical announcement directs commercial self-deployment users to a separate commercial licence.
Next check
Do not clear a commercial proof of concept from the word research. Establish the applicable agreement with Mistral before the affected business use; this page does not verify a commercial licence or customer eligibility.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself · Pinned publisher model card
Licence passages: §3.2, §4.2, §9 Research Purposes. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Mistral AI
Source support and record
- Record id
- mistral-large2-licence
- Event date
- 2024-07-24
- Source date
- 2024-07-24
- Review status
- source-checked · 2026-09-26
- Era
- agents
Licence — source support (close paraphrase): Research and non-commercial use are under the Research License; commercial self-deployment requires a Commercial License. Location: Mistral Large 2, licensing paragraph.
source-checked milestone · 2 Aug 2024 · DeepSeek
Automatic prefix-cache billing
Decision: CostEvidence: AnnouncementAustralian access: unresolved
DeepSeek introduces a distinct billing rule for repeated prompt prefixes.
Why it is a row
Recalculate repeated-context workloads using measured cache hits instead of charging every repeated input as a miss.
What this is not
Hits are not guaranteed. Historical prices on this page have since changed; no current cost or privacy guarantee is inferred.
Primary source: DeepSeek
Source support and record
- Record id
- deepseek-cache
- Event date
- 2024-08-02
- Source date
- 2024-08-02
- Review status
- source-checked · 2026-09-26
- Era
- agents
Cost — source support (close paraphrase): Caching runs automatically for all users and billing depends on actual cache hits; only identical prefixes qualify. Location: How to Use DeepSeek API’s Caching Service.
source-checked milestone · 22 Oct 2024 · Anthropic
Computer use enters public beta
Decision: CapabilitySecondary: PlaceEvidence: early accessAustralian access: unresolved
Anthropic opens computer use to developers as an experimental beta.
Why it is a row
Evaluate a supervised screen-operation pilot where the agent can carry out interface steps on the user’s behalf.
What this is not
The source calls it error-prone. It does not establish reliable unattended operation or permission to touch client systems.
Integrations: workflow and test boundary
The 22 October 2024 announcement described API computer use in public beta: inspecting screens, moving the cursor, clicking and typing. It explicitly called the capability experimental and error-prone. That history supports asking how to supervise a screen-operation trial; it supplies no current reliability result.
Next check
Compare a supplied-input task, a fixed workflow and a dynamic agent for the actual job. Use C1 for the interface, the live C2 authority record for permitted operations, and C5 for synthetic acceptance, refusal and recovery checks.
Historical beta access is not present account entitlement or authority to use client systems. No screen-operation pilot was run; reliability, recovery success and performance remain untested and unverified.
Read the Integrations source
Passage: Introduction and Teaching Claude to navigate computers, responsibly. Checked 2026-10-07. This documentary recheck is separate from the historical event date; no implementation was tested.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- claude-computer-use
- Event date
- 2024-10-22
- Source date
- 2024-10-22
- Review status
- source-checked · 2026-09-26
- Era
- agents
Capability — source support (close paraphrase): Developers can direct Claude through the API to inspect screens, move a cursor, click and type. Location: Introduction and Teaching Claude to navigate computers.
Place — source support (close paraphrase): The computer-use API lets Claude perceive and interact with ordinary computer interfaces. Location: Introduction and Teaching Claude to navigate computers.
source-checked milestone · 25 Nov 2024 · Anthropic
Model Context Protocol opens a shared integration path
Decision: Lock-inEvidence: AnnouncementAustralian access: not applicable
Anthropic releases the MCP specification, SDKs and server implementations.
Why it is a row
Evaluate reusable connectors built against an open protocol instead of a separate integration for each AI tool.
What this is not
An open protocol does not guarantee interoperability, migration effort or the safety of a particular server.
Integrations: workflow and test boundary
The 25 November 2024 announcement introduced the MCP specification, SDKs and server implementations as a shared integration path. A protocol announcement is evidence of that publication, not a tested replacement of a particular integration.
Next check
Use C3 to name the exact host/client/server boundary, required tools and fields, permissions, state, errors and switching effort. Compare both configurations with the same synthetic fixtures and rehearse restoration before recording any bounded result.
No client/server replacement, restoration or compatibility test was run. Portability, reliability and performance are untested and unverified; a handshake or matching tool name is not evidence that the workflow and controls survived.
Read the Integrations source
Passage: Introduction and Model Context Protocol components/shared connector paragraphs. Checked 2026-10-07. This documentary recheck is separate from the historical event date; no implementation was tested.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- mcp
- Event date
- 2024-11-25
- Source date
- 2024-11-25
- Review status
- source-checked · 2026-09-26
- Era
- agents
Lock-in — source support (close paraphrase): Developers can build against a standard protocol instead of maintaining separate connectors for each data source. Location: Model Context Protocol, paragraph before Getting started.
source-checked milestone · 11 Dec 2024 · Google
Gemini adds delegated Deep Research
Decision: CapabilityEvidence: AnnouncementAustralian access: unresolved
The Gemini 2.0 announcement contains a concrete research-assistant feature.
Why it is a row
Evaluate delegating an initial research-and-reporting pass, with a person checking its sources and conclusions.
What this is not
This row concerns Deep Research, not every Gemini 2.0 prototype. Research demos such as Mariner are not presented as generally available products.
Read the decision guides
Primary source: Google
Source support and record
- Record id
- gemini-deep-research
- Event date
- 2024-12-11
- Source date
- 2024-12-11
- Review status
- source-checked · 2026-09-26
- Era
- agents
Capability — source support (close paraphrase): Deep Research explores complex topics and compiles reports on the user’s behalf and is available in Gemini Advanced that day. Location: CEO note, Deep Research paragraphs.
source-checked milestone · 26 Dec 2024 · DeepSeek
V3 announces a later pricing change
Decision: CostEvidence: AnnouncementAustralian access: unresolved
The release announces a future pricing schedule rather than an immediate permanent discount.
Why it is a row
Recalculate planned API workloads against the announced post-promotion cache-hit, cache-miss and output rates.
What this is not
These are historical announced rates, not current prices. The page’s generic open-source wording is not used as proof of a named weights licence.
Primary source: DeepSeek
Source support and record
- Record id
- deepseek-v3-pricing
- Event date
- 2024-12-26
- Source date
- 2024-12-26
- Review status
- source-checked · 2026-09-26
- Era
- agents
Cost — source support (close paraphrase): The API Pricing Update retains V2 pricing until 8 February, then lists different input-hit, input-miss and output rates. Location: API Pricing Update; dated current-page navigation.
source-checked milestone · 20 Jan 2025 · DeepSeek
R1 permits use of weights and outputs under MIT
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
The release expressly opens downstream use of the model and its outputs.
Why it is a row
Evaluate permitted use of R1 weights and outputs for a model-customisation workflow.
What this is not
Do not generalise this statement to the separate licences of every distilled model. Listed API prices and benchmarks do not supply extra decision types here.
Weights and permitted use
MIT License · DeepSeek-R1 main model; distilled checkpoints have separate lineage
The pinned main-model licence grants broad permissions, including modification, distribution and sale, while requiring the copyright and permission notices in copies or substantial portions. It disclaims warranty. The publisher’s model card separately identifies Qwen and Llama origins for distilled models.
Next check
Check the specific checkpoint. Do not label every R1-distilled or third-party quantised model MIT from the main-model file; retain its upstream rights trail and notices.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself · Pinned publisher model card
Licence passages: LICENSE grant, notice condition and warranty disclaimer; README §7. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: DeepSeek
Source support and record
- Record id
- deepseek-r1
- Event date
- 2025-01-20
- Source date
- 2025-01-20
- Review status
- source-checked · 2026-09-26
- Era
- agents
Licence — source support (close paraphrase): DeepSeek announces MIT licensing for R1 and allows API outputs to be used for fine-tuning and distillation. Location: License Update.
source-checked milestone · 5 Feb 2025 · Google DeepMind
Gemini 2.0 Flash API becomes generally available
Decision: AccessEvidence: GAAustralian access: unresolved
The page changes the API release stage for 2.0 Flash.
Why it is a row
Reassess moving an API evaluation toward production now that Flash has left its experimental access stage.
What this is not
The app rollout was the prior week. Pro remained experimental and Flash-Lite was in early access; “everyone” is not a uniform access entitlement.
Selection: identity and access
Google’s 5 February 2025 announcement made Gemini 2.0 Flash generally available via the Gemini API in AI Studio and Vertex AI.
Next check
Distinguish the dated API release from the earlier app rollout and from other models with different release stages.
The app rollout was the prior week. Pro was experimental and Flash-Lite was in public preview; each model and route needs its own record.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Opening Today paragraph. Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: Google DeepMind
Source support and record
- Record id
- gemini-2-ga
- Event date
- 2025-02-05
- Source date
- 2025-02-05
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): Updated Gemini 2.0 Flash is generally available through the Gemini API in AI Studio and Vertex AI for production applications. Location: Opening body, Today paragraph.
source-checked milestone · 25 Mar 2025 · DeepSeek
V3-0324 changes the weights licence to MIT
Decision: LicenceEvidence: Weight releaseAustralian access: not applicable
The page identifies a licence change despite unchanged API usage.
Why it is a row
Recheck the permitted deployment and modification path for V3 under its newly stated MIT licence.
What this is not
Reasoning and front-end performance claims alone would not pass this gate. The licence statement is the qualifying change.
Weights and permitted use
MIT License · DeepSeek-V3-0324
The release notice identifies the V3-0324 move to MIT. The pinned licence grants broad permissions, including use, modification, distribution and sale, with a requirement to retain its copyright and permission notices. It provides no warranty.
Next check
Tie the rights record to V3-0324 and the exact files. Do not retrospectively apply these terms to every earlier V3 release, derivative or hosted service.
Licence interpretation remains qualified. No customer agreement, Australian account eligibility, deployment suitability or performance has been verified.
Read the licence itself
Licence passages: LICENSE grant, notice condition and warranty disclaimer. Checked 2026-10-05. This current licence check is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: DeepSeek
Source support and record
- Record id
- deepseek-v3-mit
- Event date
- 2025-03-25
- Source date
- 2025-03-25
- Review status
- source-checked · 2026-09-26
- Era
- agents
Licence — source support (close paraphrase): The release says the models are now under the MIT License, as R1 already was. Location: Licence bullet.
source-checked milestone · 14 May 2025 · OpenAI
GPT-4.1 reaches paid ChatGPT accounts
Decision: AccessEvidence: AnnouncementAustralian access: unresolved
The dated release note opens a product access path.
Why it is a row
Evaluate GPT-4.1 within an existing paid chat workflow instead of requiring API integration.
What this is not
The note describes historical rollout, not current model availability.
Selection: identity and access
OpenAI’s 14 May 2025 note described GPT-4.1 access in ChatGPT for Plus, Pro and Team, with Enterprise and Edu to follow.
Next check
Retain the historical plan distinctions, then obtain current model and account documentation before evaluation.
This is historical plan-specific rollout evidence, not current model availability or entitlement.
Historical rollout is not current entitlement. No Australian account, permission for client material, privacy, portability or task performance is established by this announcement.
Read the Selection source
Passage: Releasing GPT-4.1 in ChatGPT for all paid users (May 14, 2025). Checked 2026-10-05. This recheck is separate from the historical event date and original announcement check.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- gpt41-chatgpt
- Event date
- 2025-05-14
- Source date
- 2025-05-14
- Review status
- source-checked · 2026-09-26
- Era
- agents
Access — source support (close paraphrase): Plus, Pro and Team users can select GPT-4.1 that day; Enterprise and Edu access is planned later. Location: Releasing GPT-4.1 in ChatGPT for all paid users (May 14, 2025).
source-checked milestone · 22 May 2025 · Anthropic
Claude Code reaches GA across development surfaces
Decision: PlaceSecondary: AccessEvidence: GAAustralian access: unresolved
Claude Code leaves research release.
Why it is a row
Decide where a coding agent should operate: terminal, editor or background workflow.
What this is not
IDE extensions and the GitHub example are beta; product GA does not make every integration GA.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- claude-code-ga
- Event date
- 2025-05-22
- Source date
- 2025-05-22
- Review status
- source-checked · 2026-09-26
- Era
- agents
Place — source support (close paraphrase): Claude Code works in the terminal, IDEs and background through its SDK. Location: Claude Code section.
Access — source support (close paraphrase): Claude Code is now generally available. Location: Claude Code section.
source-checked milestone · 22 May 2025 · Anthropic
Claude 4 combines reasoning and tool execution
Decision: CapabilityEvidence: AnnouncementAustralian access: unresolved
The model release opens a tool-using reasoning workflow.
Why it is a row
Evaluate tasks that require the model to alternate reasoning with tool use.
What this is not
This model decision is separate from the Claude Code product’s GA decision announced that day.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- claude4-agents
- Event date
- 2025-05-22
- Source date
- 2025-05-22
- Review status
- source-checked · 2026-09-26
- Era
- agents
Capability — source support (close paraphrase): Both Claude 4 models can use tools during extended thinking, alternating reasoning and tool use. Location: Extended thinking with tool use bullet.
source-checked milestone · 22 July 2025 · Qwen Team
Qwen Code adds a terminal agent tool
Decision: PlaceEvidence: AnnouncementAustralian access: unresolved
The Qwen3-Coder post introduces a concrete CLI application.
Why it is a row
Evaluate Qwen’s coding agent in the terminal, with repository and tool access considered separately from model selection.
What this is not
The page calls Qwen Code a research-purpose tool. Its model benchmarks do not establish safe repository access or Australian service availability.
Read the decision guides
Primary source: Qwen Team
Source support and record
- Record id
- qwen-code
- Event date
- 2025-07-22
- Source date
- 2025-07-22
- Review status
- source-checked · 2026-09-26
- Era
- agents
Place — source support (close paraphrase): Alongside Qwen3-Coder, the team open-sources Qwen Code, a command-line tool for agentic coding. Location: Opening body and Qwen Code section.
source-checked milestone · 24 Nov 2025 · Anthropic
Claude Code with Claude Opus 4.5
Decision: CapabilitySecondary: PlaceEvidence: AnnouncementAustralian access: unresolved
Anthropic released Opus 4.5 and updated Claude Code that day, describing tools for longer-running agents.
Why it is a row
Decide when a longer coding-agent run is worth allowing, and which scope, stop points and checks must come first.
What this is not
The announcement does not establish an hours-long run compared with earlier Opus models or every other flagship model. Its quoted 30-minute autonomous coding session is a customer statement, not Anthropic’s own duration finding.
Read the decision guides
Primary source: Anthropic
Source support and record
- Record id
- opus-45
- Event date
- 2025-11-24
- Source date
- 2025-11-24
- Review status
- source-checked · 2026-09-26
- Era
- agents
Capability — source support (close paraphrase): Effort control, context compaction and advanced tool use let Opus 4.5 run longer with less intervention. Location: New on the Claude Developer Platform and Product updates.
Place — source support (close paraphrase): Claude Code becomes available in the desktop app for local and remote sessions. Location: New on the Claude Developer Platform and Product updates.
source-checked milestone · 29 Jan 2026 · OpenAI
ChatGPT announces legacy-model retirement
Decision: RetirementEvidence: DeprecationAustralian access: unresolved
The announcement gives a dated ChatGPT migration boundary.
Why it is a row
Identify chat workflows needing migration before the announced 13 February retirement.
What this is not
GPT-5 Instant and Thinking were previously announced for retirement. This is a historical notice, not a live inventory of available models.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- chatgpt-retirement-2026-01
- Event date
- 2026-01-29
- Source date
- 2026-01-29
- Review status
- source-checked · 2026-09-26
- Era
- now
- Effective dates
- 2026-02-13
Retirement — source support (close paraphrase): GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini will leave ChatGPT on 13 February 2026; API availability is unchanged by this notice. Location: Opening paragraph.
source-checked milestone · 28 May 2026 · OpenAI
Two ChatGPT legacy-model sunset dates
Decision: RetirementEvidence: DeprecationAustralian access: unresolved
One notice sets two product-specific deadlines.
Why it is a row
Schedule migration of remaining legacy chat workflows against both deadlines.
What this is not
ChatGPT only: the notices say the API was unchanged. OpenAI’s separate Model Release Notes still say 27 June for GPT-4.5, while the ChatGPT release notes give 26 June and confirm removal on that date. The row does not independently audit o3 completion or current model availability.
Read the decision guides
Primary source: OpenAI
Source support and record
- Record id
- chatgpt-retirement-2026-05
- Event date
- 2026-05-28
- Source date
- 2026-05-28
- Review status
- source-checked · 2026-09-28
- Era
- now
- Effective dates
- 2026-06-26, 2026-08-26
Retirement — source support (close paraphrase): The 28 May ChatGPT release note sets GPT-4.5’s retirement for 26 June and o3’s for 26 August 2026, with no API change; its 26 June update confirms GPT-4.5 was no longer available in ChatGPT. Location: Retiring OpenAI o3 and GPT-4.5 (May 28, 2026); Retiring GPT-4.5 (June 26, 2026).
reviewed milestone · 15 Sept 2026 · Google
Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking
Decision: CapabilityEvidence: AnnouncementAustralian access: unresolved
Google announced two variants with different emphases on conversational efficiency and more involved reasoning.
Why it is a row
This distinction can help readers separate voice-interface requirements from model behaviour, but the announcement does not prove availability for Australian accounts. Exact API model identifiers, eligible regions and account tiers remain unverified, and no performance claim is adopted here.
What this is not
The page says it was updated on 17 September. That later date is revision metadata for the same announcement, not evidence of a second release or, by itself, a substantive claim change.
Read the decision guides
Primary source: Google
Source support and record
- Record id
- gemini
- Event date
- 2026-09-15
- Source date
- 2026-09-15
- Review status
- reviewed · 2026-09-22
- Era
- now
- Revision
- 2026-09-17 · same event
Capability — source support (close paraphrase): The source describes variants with different emphases on conversational efficiency and more involved reasoning. Location: Retained reviewed source reading of 22 September 2026.
source-checked milestone · 15 Sept 2026 · TypeSafe
Jev early access announced
Decision: AccessEvidence: early accessAustralian access: unresolved
TypeSafe announced early access for Jev on 15 September 2026. Its later no-waitlist announcement is recorded as an access revision on this event.
Why it is a row
Records TypeSafe’s early-access announcement for Jev. It raises a documented bounded-decision interface to consider, with suitability and current account eligibility still unverified.
What this is not
Day-only evidence stays day-only. Historical public access does not establish current Australian account eligibility or an SLA. Jev accuracy in our use is unmeasured because the evaluation was dropped. No speed, saving, reliability or benchmark result is adopted.
Bounded decisions: current contract and access chronology
The current System One documentation and primitive reference describe Noul, Choice and Score. This documentary check of 7 October 2026 supplies current context, not a separate primitive launch date.
OpenRouter’s Jev 1.13 listing gives 18 September 2026 for that service. The TypeSafe no-waitlist announcement is dated 20 September 2026, 21:30:43 UTC. The warm revision mark records that public-access announcement; it does not establish current eligibility or an SLA.
Jev accuracy in our use is unmeasured because the evaluation was dropped. Examples are possibilities, and no benchmark or performance result is supplied.
Read the decision guides
Primary source: TypeSafe
Source support and record
- Record id
- jev-early-access
- Event date
- 2026-09-15
- Source date
- 2026-09-15
- Review status
- source-checked · 2026-10-07
- Era
- now
- Revision
- 2026-09-20 · same event
Access — source support (close paraphrase): The founder-authored announcement introduces Jev and invites early-access admission. Location: Announcement heading date: 15 September 2026; founder byline and early-access admission paragraph.
No milestones in this view. An empty filter is not evidence that nothing happened.