Changes to act on
Dated changes in AI tools and models, who they affect, and what to do before they land. Edits to this site are in What's new.
No changes from are listed right now. Show all makers
Budget for DeepSeek Search and Fetch calls: in effect
Updated : The change and what to do now say per-request billing started on October 31, 2026.
The change
Search and Fetch calls via server-side tools in Claude Code, Codex and DeepSeek Harness were free for DeepSeek API users. Per-request billing started October 31, 2026, at 16:00 UTC.
Who it affects
DeepSeek API users whose Claude Code, Codex or DeepSeek Harness sessions call Search or Fetch.
What to do
Count successful Search and Fetch requests one run makes, multiply by how often it runs, and add that to your forecast. Then pick the number that should slow or stop it.
Read the Cost forecasting card →
Source: DeepSeek, DeepSeek-V4.1-Flash-1017: Wish-Driven Programming Made Real · DeepSeek, Search & Fetch · Last checked
Pin xAI Imagine image quality before
The change
On November 2, 2026, requests to grok-imagine-image-quality are served by grok-imagine-image-2.0 at quality low. The call still succeeds, but you get low instead of the quality you choose.
Who it affects
Builders calling grok-imagine-image-quality or grok-imagine-image-pro (which already redirects to it), not grok-imagine-image (1.0).
What to do
Before November 2, 2026, set model to grok-imagine-image-2.0 and pass quality explicitly: low is what the redirect gives you, medium adds finer detail and costs more. Compare a few outputs to choose.
Read the Model versions card →
Source: xAI, Release Notes · xAI, grok-imagine-image-quality Retirement on November 2, 2026 · Last checked
Leave five Claude models in some Bedrock Regions before
The change
On or soon after November 13, 2026, requests fail in Regions Amazon lists for each model (all five include ap-south-1): anthropic.claude-3-haiku-20240307-v1:0, anthropic.claude-3-sonnet-20240229-v1:0, anthropic.claude-3-5-sonnet-20240620-v1:0, anthropic.claude-3-5-sonnet-20241022-v2:0, anthropic.claude-3-7-sonnet-20250219-v1:0. Migration will not happen automatically.
Who it affects
Builders on these IDs in those Regions, facing higher provider-set pricing until then, per Amazon.
What to do
Before November 13, 2026, call ListFoundationModels or GetFoundationModel in your Region, read modelLifecycle, and switch your model ID to an Active Claude model on Bedrock. Run one real task on both first.
Read the Model versions card →
Source: Amazon Bedrock, Model lifecycle (Legacy) · Last checked
Move off Claude Sonnet 4.5 by
The change
Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) is scheduled to retire on the Claude API on November 30, 2026. Anthropic recommends claude-sonnet-5-5.
Who it affects
Builders who call claude-sonnet-4-5-20250929 on the Claude API or another Anthropic-operated platform.
What to do
Run one real task on claude-sonnet-4-5-20250929 and claude-sonnet-5-5 with the same prompt, then switch before November 30, 2026. Anthropic advises testing well before the retirement date.
Read the Model Selection & Routing card →
Also useful: Evaluation basics
Source: Anthropic, Model deprecations · Anthropic, Claude Platform release notes · Last checked
Copy your OpenAI evals out before
Corrected : Headline and what-to-do now say to copy evals out before the November 30, 2026 shutdown. Existing evals are read-only from October 31, 2026.
The change
Existing OpenAI evals become read-only on October 31, 2026, and the Evals dashboard and API are scheduled to shut down on November 30, 2026.
Who it affects
Builders who run or store evals or graders on OpenAI's Evals platform.
What to do
Copy your eval cases and grading rules out of OpenAI Evals and rebuild them in a tool you control before November 30, 2026. OpenAI says "graders documented for eval workflows are part of this transition."
Read the Production evaluation card →
Also useful: Evaluation basics
Source: OpenAI, Deprecations · Last checked
Move off Agent Builder by
The change
OpenAI's Agent Builder is scheduled to shut down on November 30, 2026. ChatKit remains available.
Who it affects
Builders with workflows designed in Agent Builder.
What to do
Move each workflow before November 30, 2026. OpenAI's Deprecations page points to the Agents SDK or ChatGPT Workspace Agents.
Source: OpenAI, Agent Builder · OpenAI, Deprecations · Last checked
Plan for Gemini 4 Argon price rise before
The change
Google says Gemini 4 Argon's introductory pricing runs through November 30, 2026. From December 1, 2026, input, output (including thinking tokens) and cached input rates double.
Who it affects
Builders calling gemini-4-argon on any Gemini API paid tier (Argon has no Free tier).
What to do
Before December 1, 2026, re-run your Argon cost-per-run estimate at December 1 rates on Google's pricing page. If the cost no longer fits, test a lower thinking level or cheaper model on your eval set.
Read the Cost forecasting card →
Source: Google, What's new in Gemini 4 Argon · Google, Gemini Developer API pricing · Last checked
Leave three older GPT Image models before
The change
On December 1, 2026, OpenAI removes these from the API: gpt-image-1-mini, gpt-image-1.5 (and gpt-image-1.5-2025-12-16), chatgpt-image-latest. Calls fail after that.
Who it affects
Builders generating images through the OpenAI API with those model names.
What to do
Before December 1, 2026, switch to the replacement OpenAI names (gpt-image-2.5-sunburst or gpt-image-2.5-flare) and compare a few outputs on your own prompts first.
Read the Model versions card →
Source: OpenAI, Deprecations · Last checked
Move off six GPT-5 and o3 models before
The change
On December 11, 2026, OpenAI removes from the API: gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, and o3-pro (aliases and their dated snapshots). Calls stop working; pinning a snapshot won't protect you.
Who it affects
Builders whose OpenAI API code still calls those six names or their snapshots.
What to do
Before December 11, 2026, search your code and config for these six names. Test each replacement on your saved eval set, then switch: gpt-5.6-terra for gpt-5-mini, gpt-5.6-luna for gpt-5-nano, gpt-5.6-sol for the rest (pros: reasoning.mode: pro).
Read the Model versions card →
Also useful: Model Selection & Routing
Source: OpenAI, Deprecations · Last checked
Save Gem files that won't move to skills before
The change
Google says Gems on personal accounts transition to skills and become read only on November 17, 2026, then are removed December 17, 2026. Some files will not carry over.
Who it affects
People 18 or over who use Gems with a personal Google account.
What to do
Copy each Gem's instructions into a document you control. Before December 17, 2026, download what won't move: GitHub, audio, video, archive and legacy Office files, and the largest files in Gems over 100 MB.
Set up a task once, reuse it every time →
Source: Google, About the transition from Gems to skills · Last checked
Create planned OpenAI fine-tuning jobs before
- : Creating fine-tuning jobs or training is not available to organizations that have not previously run fine-tuning.
- : Creating fine-tuning jobs is no longer available to organizations that have not run inference on a fine-tuned model in the past 60 days.
The change
From January 6, 2027, active existing OpenAI customers can't create new fine-tuning jobs. OpenAI says inference on fine-tuned models "will be disabled only when the underlying base model is deprecated."
Who it affects
Builders whose OpenAI organisation can still create fine-tuning jobs.
What to do
If you retrain on OpenAI, plan where you'll train next before January 6, 2027. If your model is fine-tuned on a snapshot OpenAI removes on January 14, 2027, plan for it to stop by then.
Read the Fine-tuning vs prompting vs RAG card →
Source: OpenAI, Deprecations · Last checked
Move off five GPT-4 series snapshots by
The change
Five dated snapshots of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave OpenAI's API on January 14, 2027. OpenAI recommends gpt-6-luna instead.
Who it affects
Builders who call those snapshots, including fine-tuned versions of them.
What to do
Test gpt-6-luna on your saved eval set before January 14, 2027. Then re-forecast cost on its current rates.
Read the Cost forecasting card →
Also useful: Model Selection & Routing
Source: OpenAI, Deprecations · Last checked
Move off four Qwen3.5 models by
The change
Alibaba Cloud says Model Studio will retire qwen3.5-plus, two of its dated snapshots and qwen3.5-397b-a17b on January 19, 2027. It recommends qwen3.8-flash instead.
Who it affects
Builders whose Model Studio code calls any of the four listed Qwen3.5 models.
What to do
List every Qwen model name your code and config send. Test qwen3.8-flash on your saved eval set before January 19, 2027, then switch on purpose.
Read the Model versions card →
Source: Alibaba Cloud, Selected Legacy Qwen3.5 Models Decommissioning Notice · Alibaba Cloud, Model decommissioning policy · Last checked
Decide whether Gemini API stores your requests
The change
Google says paid-tier Gemini API projects created on or after October 26, 2026 store Generate Content requests by default. Projects created earlier keep their current setting.
Who it affects
Builders with paid-tier Gemini API projects created on or after October 26, 2026.
What to do
Decide whether each project should keep prompts; if not, set store=false on requests or turn off Generate Content API logging. Write down which Google page you checked, and the date.
Read the Privacy & data handling card →
Source: Google, Logs and datasets · Google, Release notes · Last checked
Remove sampling parameters from Claude and Gemini calls
The change
Claude API version 2026-10-21, the new SDK default, rejects temperature, top_p and top_k. Google says Gemini 4 and 3.6+ ignore custom values and will return 400 errors later.
Who it affects
Builders who set sampling parameters in Claude or Gemini 4 or 3.6+ requests.
What to do
Remove the three parameters, then check one prompt you already send keeps its shape. A Claude request relying on a lower temperature can stay on version 2023-06-01; Anthropic says no other setting replaces it.
Also useful: Model Selection & Routing
Source: Anthropic, Versions · Anthropic, Migrating to API version 2026-10-21 · Anthropic, Claude Platform release notes · Google, Prompt design strategies · Last checked
Update Gemini Flash model strings you call
The change
Google says API requests to gemini-3.7-flash are automatically routed to gemini-3.8-flash, and requests to gemini-3.5-flash route to gemini-3.6-flash. Update your model string.
Who it affects
Builders whose Gemini API code still sends gemini-3.5-flash or gemini-3.7-flash.
What to do
List every Gemini model name your code and config send. Replace gemini-3.5-flash and gemini-3.7-flash with a snapshot you tested, then re-check answers on your eval set.
Read the Model versions card →
Source: Google, Release notes · Last checked
Window, output cap, rate link, cache rules, and shutdown date for a model ID: Model facts. Cache size and time limit: Prompt caching.