AI assistants: model, search and crawler changes, dated

The engine changes tracker lists dated changes to the models, web search and crawlers of Claude, ChatGPT, Google (Gemini, Search and AI features), Grok, Perplexity, Copilot and open models, one table per assistant, each row linked to the maker's own page.

Each row gives the date of the change, the change in our words, a link to the maker's own announcement or documentation, what the change may affect in how an assistant finds and names products, and whether we have run our tests again since.

Claude

DateChangeSourceWhat it may changeRe-test
Claude Fable 5.2 was launched, built for demanding reasoning and long-horizon agentic work.Claude release notesAnswers from memory vs search; roundup and comparison pages as sources; directory pages for little-known agent skills; general questions answered from memory; how citations are counted; page titles; FAQ and question formatsNot re-tested yet
Fast mode for Claude Opus 5.5 left research preview.Claude release notesNothing we measuredNo re-test needed
Introduced Claude Haiku 5.5 for high-volume, cost-sensitive tasks.AnthropicNothing we measuredNo re-test needed
Claude Code added Haiku 5.5 as its default Haiku model and exposed subagent type data.Claude Code releasesAnswers from memory vs search; roundup and comparison pages as sources; directory pages for little-known agent skills; general questions answered from memory; how citations are counted; page titles; FAQ and question formatsNot re-tested yet
Claude Haiku 5.5 launched, with documented compatibility changes from Haiku 4.5.Claude release notesAnswers from memory vs search; roundup and comparison pages as sources; directory pages for little-known agent skills; general questions answered from memory; how citations are counted; page titles; FAQ and question formatsNot re-tested yet
The Models API now exposes server tool support for each model.Claude release notesNothing we measuredNo re-test needed
Anthropic launched Claude Sonnet 5.5 (claude-sonnet-5-5) on the Claude API and through Amazon Bedrock, Google Cloud and Microsoft Foundry.Claude Platform release notesAnswers from memory vs search; directory pages for little-known agent skills; general questions answered from memoryRe-tested: already in our Claude runs of and
Anthropic opened a developer portal for submitting plugins to the Claude directory, with review tracking and usage analytics after a plugin goes live.Claude apps release notesWhich directories assistants openNot re-tested yet
Introduced Claude Opus 5.5 as the first model in the Claude 5.5 family.AnthropicNothing we measuredNo re-test needed
Anthropic launched Claude Opus 5.5 (claude-opus-5-5) on the Claude API and through Amazon Bedrock, Google Cloud and Microsoft Foundry.Claude Platform release notesAnswers from memory vs search; directory pages for little-known agent skills; general questions answered from memoryRe-tested: already in our Claude runs of and

ChatGPT

DateChangeSourceWhat it may changeRe-test
GPT-Live-1.1 improves pronunciation, accent consistency, and transcription in multilingual voice conversations.OpenAI API changelogNothing we measuredNo re-test needed
The OpenAI API changelog lists GPT-6.2 Astra for demanding reasoning and coding tasks and GPT-6.2 Sol for everyday coding and agent workflows.OpenAI API changelogWhether the assistant searches by default; qualifiers and follow-up questions; assistants naming their own products; page titles; how often small brands get named; when an assistant searchesNot re-tested yet
GPT-6-Cyber is listed for investigating vulnerabilities, reproducing them in an authorized test environment, and verifying fixes.OpenAI changelogNothing we measuredNo re-test needed
GPT-6 Rosalind was released for life sciences research, with access for organizations approved for Rosalind Research.OpenAI changelogNothing we measuredNo re-test needed
OpenAI updated the regularly changing chat-latest model snapshot.OpenAI API changelogWhether the assistant searches by default; qualifiers and follow-up questions; assistants naming their own products; page titles; how often small brands get named; when an assistant searchesNot re-tested yet
OpenAI released the Decisions API beta with GPT-6 Luna.OpenAI API changelogNothing we measuredNo re-test needed
OpenAI released GPT-6.1 Sol in ChatGPT Work, Codex and the API (gpt-6.1-sol); OpenAI says it is not yet available in ChatGPT chat.OpenAI: Introducing GPT-6.1 SolNothing we measuredNo re-test needed
OpenAI released GPT-6 Sol and GPT-6 Luna in ChatGPT Work, Codex and the API; OpenAI says they are not yet available in ChatGPT chat.OpenAI: Introducing GPT-6 Sol and LunaNothing we measuredNo re-test needed
OpenAI introduced GPT-6 Astra, rolling out to paid ChatGPT plans and to the API as gpt-6-astra.OpenAI: GPT-6 AstraWhether the assistant searches by default; assistants naming their own products; when an assistant searchesNot re-tested yet

Google (Gemini, Search, AI Overviews, AI Mode)

DateChangeSourceWhat it may changeRe-test
Gemini 4 Hydrogen became available for developers to build with.Google Gemini blogWhether the assistant searches by default; small directories on narrow skill questions; assistants naming their own products; when an assistant searchesNot re-tested yet
Google Search now uses Gemini 4 Hydrogen and it is available globally for all users.Google Search blogWhether the assistant searches by default; small directories on narrow skill questions; assistants naming their own products; when an assistant searchesNot re-tested yet
Google Search Central published guidance on crawler traffic after moving to a CDN and on blanket bot protection blocking crawlers.Google Search Central blogNothing we measuredNo re-test needed
Gemini 4 Carbon was made available to developers, enterprises, scientists, and consumers.Google blogWhether the assistant searches by default; small directories on narrow skill questions; assistants naming their own products; when an assistant searchesNot re-tested yet
Gemini Omni 1.2 Flash is rolling out and generally available across the Gemini API and Gemini for business.Gemini blogNothing we measuredNo re-test needed
Google updated AI Mode and AI Overviews with clearer inline links and more source context.Google blogNothing we measuredNo re-test needed
Google added Gemini 4 Argon to the Gemini app.Google blogNothing we measuredNo re-test needed
Gemini Nano Banana 2.1 became generally available, while gemini-3.1-flash-image was deprecated.Gemini API changelogNothing we measuredNo re-test needed
Google updated its guidance on generative AI content: AI-written text, titles, meta descriptions, structured data and alt text should be fact-checked by a person before publishing, with pointers to the rater guidelines on scaled content.Google Search Central documentationCrawl, index and snippet rulesNot re-tested yet
Google announced Gemini 4 Argon, released first only to vetted security teams in its Fairwind Program, not yet to the Gemini app or Search.Google blog: Gemini 4 ArgonNothing we measuredNo re-test needed
Google added skills to Gemini chat globally: reusable instructions a user saves and calls with a slash and the skill name; Workspace business and education customers follow in the coming weeks.Google: Let skills in Gemini tackle your most repetitive tasksSmall directories on narrow skill questions; assistants naming their own productsNot re-tested yet
Google Search Console added a multimodal search filter (Lens, Circle to Search, image uploads) to the Search results report and the Generative AI features report.Google Search Central blogNothing we measuredNo re-test needed
Google started the September 2026 spam update for Search, global and in all languages, with a rollout of up to two weeks.Google Search Status DashboardCrawl, index and snippet rulesNot re-tested yet

Grok

DateChangeSourceWhat it may changeRe-test
xAI introduced Grok 4.7 in the Grok API, Grok Build and Cursor; the announcement does not mention web search or citations.xAI news: Introducing Grok 4.7Nothing we measuredNo re-test needed

Perplexity

DateChangeSourceWhat it may changeRe-test
Perplexity replaced the retrieval and ranking engine behind its search with Photon, an in-house engine that also runs a new fast preset of its Search API.Perplexity blog: PhotonWhether the assistant searches by default; being cited vs being used; when an assistant searchesNot re-tested yet

Copilot

DateChangeSourceWhat it may changeRe-test
MAI Code 2 and MAI Code 2 Flash, two new Microsoft AI coding models, are available in GitHub Copilot to eligible GitHub Copilot subscribers on annual plans.GitHub Copilot changelogNothing we measuredNo re-test needed
Bing offers a preview browser extension and an -ai search operator to get results without AI-generated answers.Bing Search BlogNothing we measuredNo re-test needed
GitHub Copilot deprecated six models and identified supported alternatives.GitHub Copilot changelogAI text files and special markup; instructions hidden in listings; crawl, index and snippet rulesNot re-tested yet
Bing Webmaster Tools added the Topics Portfolio beta to its AI Performance report.Bing Webmaster blogNothing we measuredNo re-test needed
Claude Haiku 5.5 became available as a selectable model in GitHub Copilot.GitHub ChangelogNothing we measuredNo re-test needed
GitHub made GPT-6.1 Sol generally available in GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed
GitHub made Claude Sonnet 5.5 generally available in GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed
GitHub added Claude Opus 5.5 to GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed
GitHub added GPT-6 Sol and GPT-6 Luna to GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed
GitHub started rolling out Grok 4.7 in GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed
GitHub made GPT-6 Astra generally available in GitHub Copilot.GitHub changelogNothing we measuredNo re-test needed

Open models (DeepSeek, Qwen, Kimi, GLM, Llama, Mistral)

DateChangeSourceWhat it may changeRe-test
DeepSeek released DeepSeek-V4.1-Flash-1017 on its website, app and API.DeepSeek API newsNothing we measuredNo re-test needed
Mistral released the Timestral family of time series foundation models.Mistral ResearchNothing we measuredNo re-test needed
Launched a public preview of Mistral Large 4.Mistral blogNothing we measuredNo re-test needed
DeepSeek released DeepSeek-V4.1-Flash with open weights on Hugging Face and made it the model behind its API; we do not measure DeepSeek's assistant.DeepSeek API newsNothing we measuredNo re-test needed