Skip to main content
AI News HomeChange Radar

Official cutoffs · 12 monitored platforms

AI model shutdown and API deprecation deadlines

A date-ordered calendar of breaking AI platform changes. Every deadline links to the exact affected identifiers, observable symptoms, replacement, and official provider source.

Subscribe to calendar (.ics)Download registry (.csv)

Verified July 26, 2026

Upcoming deadlines

Ordered by the platform's announced shutdown or effective date. Verify again before making a production cutover.

8 days left

Claude Opus 4.1 retires on August 5, 2026

Replace the retired model ID with claude-opus-4-8 and regression-test model behavior before the cutoff.

Anthropic · Claude API13 days left

Gemini embedding-2-preview shuts down on August 10, 2026

Move embedding requests to the generally available gemini-embedding-2 model.

Google · Gemini API20 days left

Imagen 4 API models shut down on August 17, 2026

Migrate image-generation workflows to gemini-3.1-flash-image using Google's current image-generation guidance.

Google · Gemini API29 days left

OpenAI Assistants API shuts down on August 26, 2026

Use the official Assistants-to-Conversations migration guide to move state and execution to Responses and Conversations.

OpenAI · OpenAI API62 days left

Four legacy OpenAI models shut down on September 28, 2026

Choose gpt-5.4-mini or gpt-5-mini based on workload requirements, then migrate and test the integration.

OpenAI · OpenAI API

Breaking behavior

Already in effect

These changes can affect current requests now, even when a provider has not announced a future shutdown date.

AnthropicClaude APIhigh

Claude Opus 4.7 and later reject non-default sampling parameters

Anthropic has deprecated temperature, top_p, and top_k on Claude Opus 4.7 and later; non-default values return an HTTP 400 error.

breaking behaviorStop sending temperature, top_p, and top_k for affected models. Express tone, variability, and formatting requirements in the system prompt instead.
AnthropicClaude APIhigh

Claude Sonnet 5 changes thinking defaults and token counts

Claude Sonnet 5 enables adaptive thinking by default, removes manual thinking budgets, rejects non-default sampling parameters, and tokenizes the same text into roughly 30% more tokens.

breaking behaviorRemove manual thinking budgets and non-default sampling parameters, rely on adaptive thinking, and use Anthropic's token-counting API to recalibrate limits and cost controls.
GoogleGemini APIhigh

Gemini deprecates temperature, top_p, and top_k

Gemini 3.6 Flash and 3.5 Flash-Lite ignore the sampling parameters; Google says future model generations will return HTTP 400 when they are supplied.

parameter deprecationStrip temperature, top_p, and top_k from Gemini requests and encode response constraints in a clear system instruction.

How to use this deadline calendar

Search your repositories and deployment configuration for every affected model ID, endpoint, SDK symbol, and request parameter. Migrate in staging before the listed date, then regression-test response shape, tool use, token consumption, latency, and cost.

Dates are recorded as calendar dates exactly as providers publish them. When a provider does not state a stable time and timezone, plan to finish the migration before that date begins.