LLM Changelog

Learn how Microsoft protects your organization's data.
Data protection and securityOverview of Microsoft's AI security protections for your organizational data.Security for Microsoft 365 Copilot ChatSecurity for Microsoft 365 CopilotAudit logging and retentionLearn about auditing, eDiscovery, legal hold, and data retention controls.Audit capabilities in Microsoft 365 CopilotResponsible AIMicrosoft's commitment to ethical AI principles in Copilot.Microsoft's responsible AI principles

Continue your Copilot and agents learning with these Microsoft resources.
Copilot trainingAccess videos, tutorials, and learning paths for Microsoft 365 Copilot.Explore scenariosSee real-world scenarios that demonstrate how Copilot can deliver business value.AI and Copilot videosVideos about how to get the most out of Copilot.Copilot Prompt GalleryExplore examples of how to interact with Copilot across apps using natural language prompts.

- Released the new [Built-in Tools and Function Calling Combination](https://ai.google.dev/gemini-api/docs/tool-combination) feature, making it possible to use Gemini's built-in tools alongside custom function calling tools in a single API call.
- [Grounding with Google Maps](https://ai.google.dev/gemini-api/docs/maps-grounding#supported_models) is now supported for Gemini 3 models going forward.

We’re rolling out GPT-5.4 mini in ChatGPT. GPT-5.4 mini is available to Free and Go users via the “Thinking” feature in the + menu. For all other users, GPT-5.4 mini is available as a rate limit fallback for GPT-5.4 Thinking.For Plus, Pro, and other paid users, GPT-5.4 mini will be used as a fallback for GPT-5.4 Thinking when rate limits are reached, helping with continued access to reasoning capabilities during high usage. Enterprise customers will retain the option to default Auto routing to GPT-5.4 mini, if preferred.GPT-5.4 mini will not appear as a selectable model in the model picker, and GPT-5 Thinking mini will be retired as a selectable option in 30 days. Learn more in ourblog post.

We’ve simplified the model picker to make it easier to choose the right level of reasoning.Users on Plus, Pro, Business, Enterprise, and Edu accounts will see a set of model options depending on their plan, including:
Instant– fast responses for everyday questionsThinking– deeper reasoning for more complex tasksPro– the most advanced reasoning models
The Auto functionality can be accessed by clickingConfigurein the model picker menu and choosing "Auto-switch to Thinking.” This setting will automatically be on if your last chat was set to Auto.
Configure more advanced settings
If you want more control, clickConfigureto:
Turnautomatic switchingbetween Instant and Thinking on or offAccesslegacy modelsSetthinking effortwhen Thinking or Pro is selected
We’ve also simplified theretry menu, making it easier to regenerate responses with Thinking or Pro models by clicking on the three dots underneath the response.

- Released `gemini-embedding-2-preview`, our first multimodal embedding model.
It supports text, image, video, audio, and PDF inputs,
mapping all modalities into a unified embedding space. To learn more, see
[Embeddings](https://ai.google.dev/gemini-api/docs/embeddings).

- Deprecation announcement: The `gemini-2.5-flash-lite-preview-09-2025` model
will be [shut down](https://ai.google.dev/gemini-api/docs/deprecations) on March 31, 2026.

Interactive learning for math and science
When you ask ChatGPT about certain topics, it can now present aninteractive visual modulethat lets you experiment with formulas and variables in real time. Adjust inputs, manipulate equations, and instantly see how changes affect graphs and outcomes—turning abstract concepts into something you can explore directly.Interactive learning launches with70+ math and science topics, including concepts like the Pythagorean theorem, ideal gas law, circle area, and lens equations.This feature builds on other learning tools in ChatGPT, likestudy modeandquizzes, designed to help students work through problems step by step and strengthen understanding.Try asking:
“Help me understand the Pythagorean theorem.”“Explain how PV = nRT works.”“How do I find the area of a circle?”
Availability:Rolling out to all logged-in ChatGPT users across consumer and business plans.
Auto top-up for flexible usage credits
Eligible ChatGPT Plus and Pro users can now turn on auto top-up for credits used with Codex and Sora. Auto top-up automatically replenishes your shared credit balance when it falls below the minimum balance you choose, using your default payment method on file to purchase only the amount needed to reach your target balance.
You can manage auto top-up from Codex Settings → Usage Dashboard. Auto top-up applies to the same shared credit balance used across supported features like Codex and Sora, and if your balance is already below your selected minimum when you turn it on, a top-up may happen right away. Learn more:Using Credits for Flexible Usage in ChatGPT (Free/Go/Plus/Pro) & Sora.

The Codex app is now available on Windows for ChatGPT plans that include Codex. The app gives users a Windows desktop surface for running multiple Codex agents in parallel, with isolated worktrees and reviewable diffs that can be edited, discarded, or turned into a pull request.Users can sign in with ChatGPT from the app and keep work moving across the app, CLI, and IDE without switching between tools for every task. Learn more:Using Codex with your ChatGPT plan.

What
: Added a new section on
specifying a preferred image with metadata
to the image SEO best practices and
Discover documentation
.
Why
: Based on feedback, we're clarifying that Google uses both schema.org markup and the
og:image
meta
tag as sources when determining image thumbnails in Google Search and
Discover.

Web
Edit messages with images.You can now edit messages that include image attachments, so you don’t have to start over to adjust your prompt or regenerate results.Open search results in a new tab.Cmd+click in the search panel now opens results in a new tab, making it easier to explore without losing your place.Faster sharing.Sharing chats is now easier and faster, with a quicker-loading share menu.Export visuals from Code Blocks.Flowcharts, diagrams, and data visualizations created in Code Blocks can now be exported as images for easy reuse.
Android
Automatic scroll to latest message.Existing conversations now automatically scroll to the bottom, so you can jump right back in.Faster sidebar actions.Sidebar interactions, such as renaming chats, are now faster and more responsive.Improved dictation visuals.Dictation now has a clearer, more streamlined interface that keeps your text visible and updates as you speak.Keyboard opens automatically when attaching files.The keyboard now opens automatically after you attach a file, helping you continue your message seamlessly.
Bug fixes
Fixed increased message streaming errors in GPT-5.2 ThinkingFixed a flicker affecting the Voice Mode button when opening a conversationSidebar actions on mobile web no longer stick while scrolling

- Launched Nano Banana 2, [Gemini 3.1 Flash Image Preview](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-flash-image-preview), a high-efficiency
model optimized for speed and high-volume use cases.

- Deprecation announcement: Gemini 3 Pro Preview (`gemini-3-pro-preview`)
will be [shut down](https://ai.google.dev/gemini-api/docs/deprecations) March 9, 2026.

- Released [Gemini 3.1 Pro Preview](https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview), our latest iteration in the new Gemini 3 series family.
- Launched a separate endpoint `gemini-3.1-pro-preview-customtools`, which is better at prioritizing custom tools, for users building with a mix of bash and tools.

Gemini 3.1 Pro: A smarter model for your most complex tasksWhat:Building on the Gemini 3 series, we’re releasing 3.1 Pro, a smarter and more capable model for complex problem-solving. 3.1 Pro is designed for tasks where a simple answer isn’t enough, taking advanced reasoning and making it useful for your hardest challenges. This improved intelligence can help in practical applications — whether you’re looking for a clear, visual explanation of a complex topic, a way to synthesize data into a single view, or a step-by-step plan bringing an ambitious creative project to life.3.1 Pro is rolling out globally to the Gemini app, with higher limits for users with Google AI Pro and Ultra plans. To access, select “Pro” in the model drop down.Why:We want to make Gemini the most helpful AI assistant for everyone. With the advanced reasoning of Gemini 3.1 Pro, you can now tackle your most complex projects with greater confidence.

We’re making an update to ChatGPT Voice that improves its ability to follow user instructions, and usage of tools like web search to provide better responses. This update applies to the version of Voice primarily used by those on the free ChatGPT plan, and by Plus users when limits are reached for the main model.

Jan 10, 2026:We lowered the Standard and Light thinking time as we observed users prefer faster responses. As part of this update, theExtendedthinking setting for GPT-5.2 was unintentionally changed to be lower which we have now fixed.
February 3, 2026:We made another small reduction to Standard thinking time based on testing.
February 4, 2026:We’re restoring theExtendedthinking level for GPT-5.2 Thinking to its prior setting, correcting the inadvertent reduction from January. Extended is now back to its prior level.We periodically adjust the default thinking time for our reasoning models. These changes are based on ongoing experiments to find the best balance between answer quality and response speed for users.Thethinking level toggleintroduced in September 2025 gives users more choice beyond Standard, allowing them to select the right thinking level for their question—whether they want lighter, faster responses or more extended reasoning when depth and accuracy matter more.Thinking time is not directly comparable across different models. Each model is tuned independently to what works best for users. We’ll continue to adjust these settings as models evolve, and will keep giving users clear controls when there are meaningful tradeoffs to choose from.

What
: While moving over the information about the default file size limits of
Google's crawlers and fetchers to the
crawler documentation
,
we also updated the
Googlebot documentation
about its own file size limits.
Why
: The original location of the default file size limits was not the most logical
place as it applies to all of Google's crawlers and fetchers, and the move enabled us to
be more precise about Googlebot's limits.

Today we’re releasing the Codex app for MacOS,  a command center for managing multiple coding agents in parallel. The app lets you run long‑horizon and background tasks, review clean diffs from isolated worktrees, see agent progress and decisions, and execute reusable skills and automations.
Available with ChatGPT plans that include Codex, with a limited‑time promo: Free and Go can try Codex with included limits, while Plus, and Pro users get 2x Codex rate limits.
Get started by downloading the macOS app, andlearn moreabout Codex.

What
: Updated the
structured data carousels (beta)
documentation and the
South African badges and refinement chips blog post
to include additional query types.
Why
: The South African carousel, badges, and refinement chips now support queries
related to food delivery, car hire, and bus booking.

We're rolling out an age prediction model on ChatGPT consumer plans to help determine whether an account likely belongs to someone under 18, so we can apply the right experience and safeguards for teens.The model looks at a combination of behavioral and account-level signals, including how long an account has existed, typical times of day when someone is active, usage patterns over time, and a user’s stated age.If an adult is placed in the under-18 experience by mistake, they can quickly verify their age with a selfie through Persona, a verification service available in Settings > Account. Users can check whether safeguards have been added and verify at any time.

- Deprecation announcement: The following models will be
[shut down](https://ai.google.dev/gemini-api/docs/deprecations) February 17, 2026:

- `gemini-2.5-flash-preview-09-25`
- `imagen-4.0-generate-preview-06-06`
- `imagen-4.0-ultra-generate-preview-06-06`
- The `gemini-2.5-flash-image-preview` has been shut down.

- Deprecation announcement: The following models will be
[shut down](https://ai.google.dev/gemini-api/docs/deprecations) February 17, 2026:

- `gemini-2.5-flash-preview-09-25`
- `imagen-4.0-generate-preview-06-06`
- `imagen-4.0-ultra-generate-preview-06-06`
- The `gemini-2.5-flash-image-preview` model has been shut down.