Changes

Releases, deprecations, shutdowns, silent redirects, API and price changes. Open a row for the summary, what breaks and the provider's own page.

Sources checked Oct 14, 2026 · 445 sources

Provider:

Platform:

Type:

When:

156 changes

Scheduled 36

Oct 14, 2026ShutdownClaude Sonnet 4 reaches end of life on Amazon BedrockAmazon Bedrock4 breaks

Bedrock moved Claude Sonnet 4 to Legacy on 14 April 2026 and public extended access began on 14 July 2026. ListFoundationModels gives the end of life as 08:00 UTC on 14 October 2026 in most Regions.

Platform: Amazon Bedrock
Model: Claude Sonnet 4
Announced: Apr 14, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.
PricingDuring public extended access, the model provider may charge higher prices until end of life.

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab); Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab)

Oct 15, 2026ReleaseGPT-6.1 Sol on Amazon BedrockAmazon Bedrock

ListFoundationModels lists GPT-6.1 Sol as active with a start-of-life time of 10:00 UTC on 15 October 2026. It follows OpenAI's first-party lifecycle terms.

Platform: Amazon Bedrock
Model: GPT-6.1 Sol

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Oct 20, 2026Price changeMistral Large 4 launch discount endsMistral API

Mistral priced mistral-large-4 at half its regular rate for two weeks from the 6 October 2026 launch. The regular input, cached-input and output prices apply after that.

Platform: Mistral API
Model: Mistral Large 4

Source: Mistral AI, Changelog (opens in a new tab); Mistral AI, API pricing (opens in a new tab)

Oct 20, 2026ShutdownGemini 2.5 Pro, Flash and Flash-Lite retire on Agent PlatformGoogle Vertex AI2 breaks

Google retires gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite on Gemini Enterprise Agent Platform. Gemini 3.8 Flash is the first replacement Google lists for each.

Platform: Google Vertex AI
Models: Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite
Replacement: Gemini 3.8 Flash

AvailabilityRequests to these model IDs on Agent Platform stop working after the retirement date.
ParametersReplacement models from Gemini 3.6 Flash on ignore temperature, top_p and top_k and reject a request that ends with a model turn.

Source: Google Cloud, Model versions and lifecycle (opens in a new tab); Google Cloud, Gemini 2.5 Pro (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab); Google, What's new in Gemini 3.6 Flash and 3.5 Flash-Lite (opens in a new tab)

Oct 21, 2026ShutdownDeepSeek, GLM, Kimi, Llama, MiniMax, Qwen and GPT OSS endpoints retire on Agent PlatformGoogle Vertex AI1 break

Google retires its managed endpoints for DeepSeek, GLM, GPT OSS 20B, Kimi K2 Thinking, Llama 3.3 70B, MiniMax M2 and Qwen3 models on Gemini Enterprise Agent Platform. Google suggests self-deploying them from Model Garden.

Platform: Google Vertex AI
Models: DeepSeek-R1, DeepSeek-V3.2, DeepSeek-V3.1, GLM-5, GLM-4.7, gpt-oss-20b, Kimi K2 Thinking, Llama 3.3 70B Instruct, MiniMax-M2, Qwen3-235B-A22B-Instruct-2507, Qwen3-Coder-480B-A35B-Instruct, Qwen3-Next-80B-A3B-Instruct, Qwen3 Next 80B A3B Thinking
Announced: Jul 21, 2026

AvailabilityAPI requests to the retired model IDs fail after the retirement date.

Source: Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab); Google Cloud, Open model deprecations (opens in a new tab)

Oct 22, 2026ShutdownDeepSeek V4 Flash 0731 leaves Together serverlessTogether AI1 break

Together removes deepseek-ai/DeepSeek-V4-Flash-0731 from serverless inference and recommends DeepSeek V4.1 Flash.

Platform: Together AI
Model: DeepSeek-V4-Flash
Replacement: DeepSeek-V4.1-Flash

AvailabilityRequests to the removed serverless model ID fail. The model can still run on an on-demand dedicated endpoint, billed per minute.

Source: Together AI, Deprecations (opens in a new tab)

Oct 22, 2026ShutdownLlama 3.3 70B Instruct Turbo leaves Together serverlessTogether AI1 break

Together removes meta-llama/Llama-3.3-70B-Instruct-Turbo from serverless inference and recommends Muse Glimmer 30B.

Platform: Together AI
Model: Llama 3.3 70B Instruct
Replacement: Muse Glimmer 30B

AvailabilityRequests to the removed serverless model ID fail. The model can still run on an on-demand dedicated endpoint, billed per minute.

Source: Together AI, Deprecations (opens in a new tab)

Oct 23, 2026ShutdownLegacy GPT and o-series snapshots shut down on the OpenAI APIOpenAI API3 breaks

Requests to gpt-3.5-turbo-0125, gpt-4-0613, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o1, o1-pro, o3-mini and o4-mini, and their aliases, fail after the shutdown date. OpenAI names GPT-5.6 Sol, Terra or Luna as the substitute for each. Fine-tuned gpt-3.5-turbo, gpt-4, gpt-4.1-nano and o4-mini models shut down the same day.

Platform: OpenAI API
Models: GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, GPT-4.1 nano, GPT-4o, o1, o1-pro, o3-mini, o4-mini
Announced: Apr 22, 2026

ParametersPro snapshots move to gpt-5.6-sol with reasoning.mode set to pro. There is no separate Pro model ID for GPT-5.6.
DefaultsGPT-5.6 models reuse reasoning across turns by default (reasoning.context all_turns); earlier models default to current_turn.
PricingGPT-5.6 bills cache writes at 1.25 times the uncached input rate. Replace prompt_cache_retention with prompt_cache_options.ttl.

Source: OpenAI, Deprecations (opens in a new tab); OpenAI, Using GPT-5.6 (opens in a new tab)

Oct 31, 2026ShutdownZ.ai GLM 5.2 retires from the Mistral APIMistral API

zai-glm-5-2 retires on 31 October 2026. Mistral names Z.ai GLM 5.3 (zai-glm-5-3) as the replacement at the same price.

Platform: Mistral API
Model: GLM-5.2
Replacement: GLM-5.3
Announced: Sep 29, 2026

Source: Mistral AI, Changelog (opens in a new tab); Mistral AI, Models Overview (opens in a new tab)

Nov 15, 2026ShutdownClaude Haiku 4.5 retires on Azure AI FoundryAzure AI Foundry1 break

Both versions of claude-haiku-4-5 on Foundry, hosted by Anthropic and hosted on Azure, retire on the same day.

Platform: Azure AI Foundry
Model: Claude Haiku 4.5

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Nov 15, 2026Shutdowncodex-mini retires on Azure AI FoundryAzure AI Foundry2 breaks

codex-mini version 2025-05-16 is deprecated and retires with no replacement listed yet.

Platform: Azure AI Foundry
Model: codex-mini-latest

AvailabilityInference requests to retired deployments return 410 Gone.
AvailabilityMicrosoft has not listed a replacement yet; it names one about 90 to 120 days before retirement. Deployments set to NoAutoUpgrade stop working at retirement.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Nov 19, 2026Shutdowno1, o3 and o4-mini models retire on Azure AI FoundryAzure AI Foundry2 breaks

o1, o1-pro, o3, o3-deep-research, o3-mini, o3-pro and o4-mini are deprecated and retire on Foundry. Microsoft lists gpt-5.6-sol as the replacement, or gpt-5.6-terra for o3-mini and o4-mini.

Platform: Azure AI Foundry
Models: o1, o1-pro, o3, o3-deep-research, o3-mini, o3-pro, o4-mini

BehaviorStandard, Global Standard and Data Zone Standard deployments are upgraded to the replacement automatically unless versionUpgradeOption is NoAutoUpgrade. Provisioned deployments are not upgraded and stop working at retirement.
AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Nov 19, 2026ShutdownGemini 3.6 Flash retires on Agent PlatformGoogle Vertex AI1 break

Google retires gemini-3.6-flash on Gemini Enterprise Agent Platform and recommends gemini-3.8-flash.

Platform: Google Vertex AI
Model: Gemini 3.6 Flash
Replacement: Gemini 3.8 Flash

BehaviorGemini 3.8 Flash can use more tokens on long, complex tasks by design.

Source: Google Cloud, Model versions and lifecycle (opens in a new tab); Google, What's new in Gemini 3.8 Flash (opens in a new tab)

Nov 24, 2026ShutdownClaude Opus 4.5 retires on Azure AI FoundryAzure AI Foundry1 break

claude-opus-4-5, hosted on Anthropic infrastructure, retires on Foundry.

Platform: Azure AI Foundry
Model: Claude Opus 4.5

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Nov 26, 2026ShutdownAI21 Jamba 1.5 Large and Mini reach end of life on Amazon BedrockAmazon Bedrock4 breaks

Bedrock moved Jamba 1.5 Large and Jamba 1.5 Mini to Legacy on 26 May 2026. Public extended access began on 26 August 2026, and requests fail after 26 November 2026.

Platform: Amazon Bedrock
Models: Jamba 1.5 Large, Jamba 1.5 Mini
Announced: May 26, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.
PricingDuring public extended access, the model provider may charge higher prices until end of life.

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab); Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab)

Nov 30, 2026ShutdownClaude Sonnet 4.5 shuts down on the Claude APIClaude API6 breaks

Requests to claude-sonnet-4-5-20250929 fail after the shutdown date. Anthropic names Claude Sonnet 5.5 as the replacement.

Platform: Claude API
Model: Claude Sonnet 4.5
Replacement: Claude Sonnet 5.5
Announced: Sep 30, 2026

ParametersClaude 4.7 and later models return a 400 error when temperature, top_p or top_k is set to a non-default value. Remove them before switching.
ParametersManual thinking budgets (thinking type "enabled" with budget_tokens) return a 400 error on Claude Sonnet 5.5. Use adaptive thinking and an effort level.
ParametersA prefilled final assistant turn returns a 400 error. Claude Sonnet 4.5 accepted one.
ParametersForced tool choice (tool_choice "any" or "tool") returns a 400 error on Claude Sonnet 5.5. Use "auto" with strict tools.
DefaultsRequests without a thinking field run with adaptive thinking on Claude Sonnet 5.5, so responses can start with thinking blocks.
TokenizerThe same text produces about 30% more tokens than on Claude Sonnet 4.5. Recount prompts and revisit max_tokens.

Source: Anthropic, Model deprecations (opens in a new tab); Anthropic, Claude Sonnet 5.5 migration guide (opens in a new tab)

Nov 30, 2026ShutdownClaude Sonnet 4.5 retires on Azure AI FoundryAzure AI Foundry1 break

claude-sonnet-4-5 retires on Foundry. Microsoft lists claude-sonnet-5-5 as the replacement.

Platform: Azure AI Foundry
Model: Claude Sonnet 4.5
Replacement: Claude Sonnet 5.5

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Dec 1, 2026Price changeGemini 4 Argon introductory price endsGemini API, Google Vertex AI1 break

Input, output and cached-input prices for gemini-4-argon double on December 1, 2026 on the Gemini API and Agent Platform.

Platform: Gemini API, Google Vertex AI
Model: Gemini 4 Argon

PricingPer-token cost doubles from December 1, 2026.

Source: Google, Gemini Developer API pricing (opens in a new tab); Google, What's new in Gemini 4 Argon (opens in a new tab); Google Cloud, Agent Platform pricing (opens in a new tab)

Dec 9, 2026ShutdownGPT-4o version 2024-05-13 retires on Azure AI FoundryAzure AI Foundry2 breaks

gpt-4o version 2024-05-13 is deprecated and retires on Foundry. Microsoft lists gpt-5.6-sol as the replacement and will upgrade Standard deployments to it in each region.

Platform: Azure AI Foundry
Model: GPT-4o
Replacement: GPT-5.6 Sol

BehaviorStandard, Global Standard and Data Zone Standard deployments are upgraded to the replacement automatically unless versionUpgradeOption is NoAutoUpgrade. Provisioned deployments are not upgraded and stop working at retirement.
AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Dec 11, 2026ShutdownFirst GPT-5 and o3 snapshots shut down on the OpenAI APIOpenAI API3 breaks

gpt-5-2025-08-07, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07, gpt-5-pro-2025-10-06, o3-2025-04-16 and o3-pro-2025-06-10 are removed. OpenAI recommends GPT-5.6 Sol for GPT-5, GPT-5 Pro, o3 and o3-pro, GPT-5.6 Terra for GPT-5 mini, and GPT-5.6 Luna for GPT-5 nano.

Platform: OpenAI API
Models: GPT-5, GPT-5 Mini, GPT-5 nano, GPT-5 Pro, o3, o3-pro
Announced: Jun 11, 2026

ParametersPro snapshots move to gpt-5.6-sol with reasoning.mode set to pro. There is no separate Pro model ID for GPT-5.6.
DefaultsGPT-5.6 models reuse reasoning across turns by default (reasoning.context all_turns); earlier models default to current_turn.
PricingGPT-5.6 bills cache writes at 1.25 times the uncached input rate. Replace prompt_cache_retention with prompt_cache_options.ttl.

Source: OpenAI, Deprecations (opens in a new tab); OpenAI, Using GPT-5.6 (opens in a new tab)

Jan 1, 2027Price changeGemini 3.8 Flash and 3.6 Flash introductory prices end on the Gemini APIGemini API1 break

Standard prices for gemini-3.8-flash and gemini-3.6-flash take effect on January 1, 2027, double the introductory rates.

Platform: Gemini API
Models: Gemini 3.8 Flash, Gemini 3.6 Flash

PricingPer-token cost doubles from January 1, 2027.

Source: Google, What's new in Gemini 3.8 Flash (opens in a new tab); Google, Gemini Developer API pricing (opens in a new tab)

Jan 1, 2027Price changeGemini 3.8 Flash and 3.7 Flash introductory prices end on Agent PlatformGoogle Vertex AI1 break

Standard prices for gemini-3.8-flash and gemini-3.7-flash take effect on January 1, 2027, double the introductory rates.

Platform: Google Vertex AI
Models: Gemini 3.8 Flash, Gemini 3.7 Flash

PricingPer-token cost doubles from January 1, 2027.

Source: Google Cloud, Agent Platform pricing (opens in a new tab)

Jan 8, 2027ShutdownClaude Opus 4.1 reaches end of life on Amazon BedrockAmazon Bedrock4 breaks

Bedrock moved Claude Opus 4.1 to Legacy on 8 July 2026. Public extended access began on 8 October 2026, and requests fail after 8 January 2027.

Platform: Amazon Bedrock
Model: Claude Opus 4.1
Announced: Jul 8, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.
PricingDuring public extended access, the model provider may charge higher prices until end of life.

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab); Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab)

Jan 14, 2027ShutdownGPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini shut down on the OpenAI APIOpenAI API3 breaks

The gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14 snapshots are removed. OpenAI names GPT-6 Luna as the replacement for all of them. Fine-tuned versions follow the same schedule.

Platform: OpenAI API
Models: GPT-4o, GPT-4o Mini, GPT-4.1, GPT-4.1 Mini
Replacement: GPT-6 Luna
Announced: Oct 14, 2026

ParametersWhen reasoning effort is not none, remove temperature, top_p and top_logprobs. Chat Completions requests must also drop logprobs.
ToolingGPT-6 Sol and GPT-6 Luna support function calling in Chat Completions only with reasoning_effort none. Use the Responses API for reasoning with tools.
PricingReplace prompt_cache_retention with prompt_cache_options.ttl set to 30m, and expect cache writes to be billed.

Source: OpenAI, Deprecations (opens in a new tab); OpenAI, Using GPT-6 (opens in a new tab)

Jan 28, 2027ShutdownGemini 3.7 Flash retires on Agent PlatformGoogle Vertex AI

Google retires gemini-3.7-flash on Gemini Enterprise Agent Platform and recommends gemini-3.8-flash. The Gemini API already routes this ID to Gemini 3.8 Flash.

Platform: Google Vertex AI
Model: Gemini 3.7 Flash
Replacement: Gemini 3.8 Flash

Source: Google Cloud, Model versions and lifecycle (opens in a new tab); Google, Gemini API release notes (opens in a new tab)

Feb 2, 2027ShutdownClaude Opus 4.6 retires on Azure AI FoundryAzure AI Foundry1 break

claude-opus-4-6, hosted on Anthropic infrastructure, retires on Foundry.

Platform: Azure AI Foundry
Model: Claude Opus 4.6

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Feb 9, 2027ShutdownGPT-5, GPT-5 mini and GPT-5 nano retire on Azure AI FoundryAzure AI Foundry2 breaks

Version 2025-08-07 of gpt-5, gpt-5-mini and gpt-5-nano retires on Foundry, two months after the same snapshots shut down on the OpenAI API.

Platform: Azure AI Foundry
Models: GPT-5, GPT-5 Mini, GPT-5 nano

AvailabilityInference requests to retired deployments return 410 Gone.
AvailabilityMicrosoft has not listed a replacement yet; it names one about 90 to 120 days before retirement. Deployments set to NoAutoUpgrade stop working at retirement.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Feb 10, 2027ShutdownClaude Sonnet 4.6 retires on Azure AI FoundryAzure AI Foundry1 break

claude-sonnet-4-6, hosted on Anthropic infrastructure, retires on Foundry.

Platform: Azure AI Foundry
Model: Claude Sonnet 4.6

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Mar 17, 2027ShutdownGPT-5-Codex retires on Azure AI FoundryAzure AI Foundry2 breaks

gpt-5-codex version 2025-09-15 retires on Foundry. It already shut down on the OpenAI API in July 2026.

Platform: Azure AI Foundry
Model: GPT-5-Codex

AvailabilityInference requests to retired deployments return 410 Gone.
AvailabilityMicrosoft has not listed a replacement yet; it names one about 90 to 120 days before retirement. Deployments set to NoAutoUpgrade stop working at retirement.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Mar 30, 2027Shutdown26 open-weight models reach end of life on Amazon BedrockAmazon Bedrock3 breaks

Bedrock moved 26 open-weight models from DeepSeek, Google, Meta, MiniMax, Mistral AI, Moonshot AI, NVIDIA, Qwen and Z.ai to Legacy on 30 September 2026. Requests to them fail after 30 March 2027. On 14 October 2026 ListFoundationModels still reported them as active.

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.

Source: Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab); Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Apr 1, 2027ShutdownGPT-5.1, GPT-5.3-Codex and GPT-5.4 nano shut down on the OpenAI APIOpenAI API2 breaks

OpenAI removes gpt-5.1, gpt-5.3-codex and gpt-5.4-nano. It recommends GPT-6 Sol for GPT-5.1 and GPT-5.3-Codex, and GPT-6 Luna for GPT-5.4 nano.

Platform: OpenAI API
Models: GPT-5.1, GPT-5.3-Codex, GPT-5.4 nano
Announced: Oct 1, 2026

ParametersWhen reasoning effort is not none, remove temperature, top_p and top_logprobs. Chat Completions requests must also drop logprobs.
ToolingGPT-6 Sol and GPT-6 Luna support function calling in Chat Completions only with reasoning_effort none. Use the Responses API for reasoning with tools.

Source: OpenAI, Deprecations (opens in a new tab); OpenAI, Using GPT-6 (opens in a new tab)

Apr 6, 2027ShutdownClaude Opus 4.7 retires on Azure AI FoundryAzure AI Foundry1 break

claude-opus-4-7, hosted on Anthropic infrastructure, retires on Foundry.

Platform: Azure AI Foundry
Model: Claude Opus 4.7

AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Apr 7, 2027ShutdownGPT-5 Pro retires on Azure AI FoundryAzure AI Foundry2 breaks

gpt-5-pro version 2025-10-06 retires on Foundry, four months after the same snapshot shuts down on the OpenAI API.

Platform: Azure AI Foundry
Model: GPT-5 Pro

AvailabilityInference requests to retired deployments return 410 Gone.
AvailabilityMicrosoft has not listed a replacement yet; it names one about 90 to 120 days before retirement. Deployments set to NoAutoUpgrade stop working at retirement.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

Apr 8, 2027ShutdownClaude Sonnet 4.5 reaches end of life on Amazon BedrockAmazon Bedrock4 breaks

Bedrock moved Claude Sonnet 4.5 to Legacy on 8 October 2026. Public extended access starts on 8 January 2027, and requests fail after 8 April 2027.

Platform: Amazon Bedrock
Model: Claude Sonnet 4.5
Announced: Oct 8, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.
PricingDuring public extended access, the model provider may charge higher prices until end of life.

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab); Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab)

Apr 14, 2027ShutdownGPT-4.1 and remaining GPT-4o versions retire on Azure AI FoundryAzure AI Foundry2 breaks

gpt-4.1, gpt-4.1-mini and gpt-4.1-nano (2025-04-14), gpt-4o (2024-08-06 and 2024-11-20) and gpt-4o-mini (2024-07-18) retire on Foundry, three months after OpenAI's own shutdown of these models. Microsoft lists gpt-5.1 as the replacement for the two gpt-4o versions. Fine-tuned deployments of these models run until October 2027.

Platform: Azure AI Foundry
Models: GPT-4.1, GPT-4.1 Mini, GPT-4.1 nano, GPT-4o, GPT-4o Mini

BehaviorStandard, Global Standard and Data Zone Standard deployments are upgraded to the replacement automatically unless versionUpgradeOption is NoAutoUpgrade. Provisioned deployments are not upgraded and stop working at retirement.
AvailabilityInference requests to retired deployments return 410 Gone.

Source: Microsoft, Model retirement schedule - Microsoft Foundry (opens in a new tab); Microsoft, Microsoft Foundry Models lifecycle and support policy (opens in a new tab)

May 7, 2027 or laterShutdownGemini 3.1 Flash-Lite can shut down from May 7, 2027 on the Gemini APIGemini API

Google lists May 7, 2027 as the earliest shutdown date for gemini-3.1-flash-lite and recommends gemini-3.5-flash-lite.

Platform: Gemini API
Model: Gemini 3.1 Flash-Lite
Replacement: Gemini 3.5 Flash-Lite

Source: Google, Gemini deprecations (opens in a new tab)

Already in effect 120

Oct 13, 2026ReleaseGemini 4 Argon releasedGemini API, Google Vertex AI4 breaks

Google released Gemini 4 Argon as generally available on the Gemini API and Gemini Enterprise Agent Platform, with a unified 1M token limit for input and output.

Platform: Gemini API, Google Vertex AI
Model: Gemini 4 Argon

ToolingExplicit context caching is not supported; remove calls that create or reference cached content. Implicit caching stays on.
Parameterstemperature, top_p and top_k are ignored.
DefaultsThe model can spend more of max_output_tokens on thinking; Google recommends a limit of at least 32,768 tokens.
PricingThe introductory price ends on November 30, 2026, and the standard price doubles it.

Source: Google, Gemini API release notes (opens in a new tab); Google, What's new in Gemini 4 Argon (opens in a new tab); Google, Gemini Developer API pricing (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Oct 13, 2026ReleaseGemini 4 Argon Cyber Permissive released for allowlisted customersGoogle Vertex AI

Google released Gemini 4 Argon Cyber Permissive, a version of Gemini 4 Argon for defensive cybersecurity work, behind the Fairwind Program allowlist.

Platform: Google Vertex AI
Model: Gemini 4 Argon Cyber Permissive

Source: Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab); Google Cloud, Gemini 4 Argon (opens in a new tab)

Oct 8, 2026RedirectGemini 3.5 Flash requests now go to Gemini 3.6 FlashGemini API1 break

Google deprecated gemini-3.5-flash on the Gemini API and routes its requests to gemini-3.6-flash without an error. Responses now come from a different model.

Platform: Gemini API
Model: Gemini 3.5 Flash
Replacement: Gemini 3.6 Flash
Announced: Oct 8, 2026

ParametersGemini 3.6 Flash ignores temperature, top_p and top_k, and returns a 400 error when the last turn is a prefilled model turn.

Source: Google, Gemini API release notes (opens in a new tab); Google, Gemini deprecations (opens in a new tab); Google, What's new in Gemini 3.6 Flash and 3.5 Flash-Lite (opens in a new tab)

Oct 8, 2026RedirectGemini 3.7 Flash requests now go to Gemini 3.8 FlashGemini API1 break

Google deprecated gemini-3.7-flash on the Gemini API and routes its requests to gemini-3.8-flash without an error. Responses now come from a different model.

Platform: Gemini API
Model: Gemini 3.7 Flash
Replacement: Gemini 3.8 Flash
Announced: Oct 8, 2026

BehaviorOutputs, token use and latency follow Gemini 3.8 Flash, which can use more tokens on long, complex tasks by design.

Source: Google, Gemini API release notes (opens in a new tab); Google, Gemini deprecations (opens in a new tab); Google, What's new in Gemini 3.8 Flash (opens in a new tab)

Oct 7, 2026ReleaseClaude Haiku 5.5 releasedClaude API, Google Vertex AI4 breaks

Anthropic released Claude Haiku 5.5 with a 1M token context window and prices that rise for prompts over 100,000 tokens. Google Cloud lists the same release date.

Platform: Claude API, Google Vertex AI
Model: Claude Haiku 5.5

ParametersThinking type "enabled" with budget_tokens returns a 400 error. Use adaptive thinking and an effort level.
ParametersNon-default temperature or top_p values, any top_k value, and assistant prefill return a 400 error.
TokenizerThe same text produces about 30% more tokens than on Claude Haiku 4.5.
PricingRequests with prompts over 100,000 tokens pay higher prices for the whole request.

Source: Anthropic, Claude Haiku 5.5 (opens in a new tab); Anthropic, Claude Haiku 5.5 migration guide (opens in a new tab); Anthropic, Pricing (opens in a new tab); Google Cloud, Claude Haiku 5.5 on Google Cloud (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Oct 7, 2026ReleaseClaude Haiku 5.5 available on Amazon BedrockAmazon Bedrock

Bedrock added Claude Haiku 5.5 as anthropic.claude-haiku-5-5.

Platform: Amazon Bedrock
Model: Claude Haiku 5.5

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Oct 7, 2026API changechat-latest updated againOpenAI API1 break

OpenAI updated the snapshot behind chat-latest to the latest model available in ChatGPT for Plus, Pro, Business and Enterprise users, and recommends the GPT-6 family for API work.

Platform: OpenAI API
Model: Chat Latest

BehaviorOutputs can change without a model ID change.

Source: OpenAI, Changelog (opens in a new tab)

Oct 6, 2026ReleaseMistral Large 4 in Public Preview on the Mistral APIMistral API

Mistral released Mistral Large 4 (mistral-large-4) in Public Preview: a mixture-of-experts model with a 1M-token context window. Mistral says open weights are coming soon. mistral-large-latest still points to Mistral Large 3.

Platform: Mistral API
Model: Mistral Large 4

Source: Mistral AI, Changelog (opens in a new tab); Mistral AI, Mistral Large 4 model card (opens in a new tab); Mistral AI, Mistral Large 3 model card (opens in a new tab)

Oct 1, 2026Price changeDeepSeek V4.1 Flash serverless prices change on FireworksFireworks AI1 break

Standard and Priority serverless rates for DeepSeek V4.1 Flash changed: uncached input and output prices rose and the cached input price fell.

Platform: Fireworks AI
Model: DeepSeek-V4.1-Flash

PricingServerless usage costs more per uncached input and output token. Dedicated deployments and Reserved Throughput are unaffected.

Source: Fireworks AI, Changelog (opens in a new tab)

Sep 30, 2026ShutdownLeanstral 1.5 retires from the Mistral APIMistral API

Mistral said at release on 30 June 2026 that the Labs model labs-leanstral-1-5 would retire on 30 September 2026, and marked it deprecated on 29 September. Mistral names no replacement.

Platform: Mistral API
Model: Leanstral 1.5
Announced: Jun 30, 2026

Source: Mistral AI, Changelog (opens in a new tab); Mistral AI, Models Overview (opens in a new tab)

Sep 29, 2026ReleaseGLM 5.3 available on Amazon BedrockAmazon Bedrock

Bedrock added GLM 5.3 as zai.glm-5.3, with at least 45 days of notice before any end of life.

Platform: Amazon Bedrock
Model: GLM-5.3

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 29, 2026ReleaseGPT-6.1 Sol released on the OpenAI APIOpenAI API2 breaks

OpenAI released GPT-6.1 Sol for complex coding and professional work at a lower cost than GPT-6 Astra. Multi-agent delegation is in beta.

Platform: OpenAI API
Model: GPT-6.1 Sol

ParametersGPT-6.1 Sol does not support the none or minimal reasoning efforts; the default is medium.
ToolingUse the Responses API for tool calling. Chat Completions supports GPT-6.1 Sol only for requests without tools.

Source: OpenAI, Changelog (opens in a new tab); OpenAI, Using GPT-6 (opens in a new tab)

Sep 28, 2026ReleaseClaude Sonnet 5.5 releasedClaude API, Google Vertex AI3 breaks

Anthropic released Claude Sonnet 5.5 at the Claude Sonnet 5 price, with cache reads at half the earlier rate. Google Cloud lists the same release date.

Platform: Claude API, Google Vertex AI
Model: Claude Sonnet 5.5

DefaultsRequests without a thinking field now run with adaptive thinking.
ParametersThinking type "disabled" returns a 400 error. Send thinking type "between_tools" to turn off up-front thinking.
ParametersForced tool choice (tool_choice "any" or "tool") returns a 400 error.

Source: Anthropic, Claude Sonnet 5.5 (opens in a new tab); Anthropic, Claude Sonnet 5.5 migration guide (opens in a new tab); Google Cloud, Claude Sonnet 5.5 on Google Cloud (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Sep 28, 2026ReleaseClaude Sonnet 5.5 available on Amazon BedrockAmazon Bedrock

Bedrock added Claude Sonnet 5.5 as anthropic.claude-sonnet-5-5.

Platform: Amazon Bedrock
Model: Claude Sonnet 5.5

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 28, 2026ReleaseGrok 4.7 available on Amazon BedrockAmazon Bedrock

Bedrock added Grok 4.7 as xai.grok-4.7, served through US and global cross-Region inference.

Platform: Amazon Bedrock
Model: Grok 4.7

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 28, 2026ReleaseZ.ai GLM 5.3 generally available on the Mistral APIMistral API

zai-glm-5-3 reached General Availability on the Mistral API. The zai-glm-5 and zai-glm-latest aliases point to it.

Platform: Mistral API
Model: GLM-5.3

Source: Mistral AI, Changelog (opens in a new tab); Mistral AI, Z.ai GLM 5.3 model card (opens in a new tab)

Sep 28, 2026ShutdownGPT-3.5 Turbo Instruct, babbage-002, davinci-002 and gpt-3.5-turbo-1106 shut downOpenAI API1 break

OpenAI removed gpt-3.5-turbo-instruct, babbage-002, davinci-002 and the gpt-3.5-turbo-1106 snapshot. It recommends GPT-5.6 Terra.

Platform: OpenAI API
Models: GPT-3.5 Turbo Instruct, babbage-002, davinci-002, GPT-3.5 Turbo
Replacement: GPT-5.6 Terra
Announced: Sep 26, 2025

Toolingbabbage-002 and davinci-002 used the legacy Completions endpoint. GPT-5.6 Terra does not support v1/completions.

Source: OpenAI, Deprecations (opens in a new tab); OpenAI, GPT-5.6 Terra model page (opens in a new tab)

Sep 25, 2026ShutdownFireworks removes older DeepSeek, GLM, Muse and Kimi models from serverlessFireworks AI1 break

DeepSeek V4 Pro (0813), DeepSeek V4 Flash (0731), DeepSeek V4 Flash Vision Exp, Muse Glimmer 30B, Kimi K2.6, Kimi K2.7 Code and GLM 5.2 left serverless, including their Fast and US-only endpoints. Fireworks named DeepSeek V4.1 Flash, NVIDIA Nemotron 3.5 Lightning 30B A3B, GLM 5.3 and Kimi K3 as migration targets.

Platform: Fireworks AI
Models: DeepSeek-V4-Pro, DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, Muse Glimmer 30B, Kimi K2.6, Kimi K2.7 Code, GLM-5.2
Announced: Sep 12, 2026

AvailabilityServerless requests to these model IDs stop working after the date. Dedicated deployments are unaffected.

Source: Fireworks AI, Changelog (opens in a new tab)

Sep 25, 2026API changeImage encoding fix changes GPT-6 Sol and Luna image resultsOpenAI API1 break

OpenAI fixed a bug in image encoding that degraded image understanding in GPT-6 Sol and GPT-6 Luna, in the API and Codex, including computer use.

Platform: OpenAI API
Models: GPT-6 Sol, GPT-6 Luna

BehaviorResults on image inputs change without a model ID change. OpenAI recommends rerunning evaluations and retrying workflows affected by the issue.

Source: OpenAI, Changelog (opens in a new tab)

Sep 23, 2026ReleaseFireworks releases Ember-1 as a serverless research previewFireworks AI

Ember-1 is a Fireworks model built on Kimi K3 that produces shorter reasoning traces. It is served as a Research Preview alongside Kimi K3.

Platform: Fireworks AI
Model: Ember-1

Source: Fireworks AI, Introducing Ember-1 (opens in a new tab)

Sep 22, 2026ReleaseClaude Opus 5.5 releasedClaude API, Google Vertex AI4 breaks

Anthropic released Claude Opus 5.5 at a lower per-token price than Claude Opus 5. Google Cloud lists the same release date.

Platform: Claude API, Google Vertex AI
Model: Claude Opus 5.5

ParametersThinking cannot be disabled: thinking type "disabled" and manual budget_tokens both return a 400 error.
ParametersForced tool choice (tool_choice "any" or "tool") returns a 400 error.
DefaultsThe default effort is medium, where Claude Opus 5 defaulted to high. Set effort explicitly.
ToolingOn the Claude API and Google Cloud, computer use needs the computer_toolset_20260801 toolset; computer_20251124 is rejected.

Source: Anthropic, Claude Opus 5.5 (opens in a new tab); Anthropic, Claude Opus 5.5 migration guide (opens in a new tab); Google Cloud, Claude Opus 5.5 on Google Cloud (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Sep 22, 2026ReleaseClaude Opus 5.5 available on Amazon BedrockAmazon Bedrock

Bedrock added Claude Opus 5.5 as anthropic.claude-opus-5-5.

Platform: Amazon Bedrock
Model: Claude Opus 5.5

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 22, 2026ReleaseGPT-6 Luna and GPT-6 Sol available on Amazon BedrockAmazon Bedrock

Bedrock added GPT-6 Luna and GPT-6 Sol.

Platform: Amazon Bedrock
Models: GPT-6 Luna, GPT-6 Sol

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 22, 2026ReleaseGPT-6 Sol and GPT-6 Luna available on Azure AI FoundryAzure AI Foundry

Microsoft added GPT-6 Sol and GPT-6 Luna to the generally available lineup in Foundry. Standard deployments are available in Global regions and US and EU Data Zones, and Provisioned Throughput is offered for Astra and Sol.

Platform: Azure AI Foundry
Models: GPT-6 Sol, GPT-6 Luna

Source: Microsoft, GPT-6 Astra, Sol, and Luna in Microsoft Foundry (opens in a new tab)

Sep 22, 2026ReleaseGPT-6 Sol and GPT-6 Luna released on the OpenAI APIOpenAI API2 breaks

OpenAI released GPT-6 Sol and GPT-6 Luna, reasoning models that take text and image input and return text, in the Responses and Chat Completions APIs.

Platform: OpenAI API
Models: GPT-6 Sol, GPT-6 Luna

ParametersWhen reasoning effort is not none, remove temperature, top_p and top_logprobs. Chat Completions requests must also drop logprobs.
ToolingGPT-6 Sol and GPT-6 Luna support function calling in Chat Completions only with reasoning_effort none. Use the Responses API for reasoning with tools.

Source: OpenAI, Changelog (opens in a new tab); OpenAI, Using GPT-6 (opens in a new tab)

Sep 22, 2026Price changeLower serverless prices for Qwen3.7 Max and Qwen3.8 Flash on TogetherTogether AI1 break

Together lowered input and output prices for Qwen/Qwen3.7-Max and Qwen/Qwen3.8-Flash from 22 September 2026. The serverless models page still lists the launch rates.

Platform: Together AI
Models: Qwen3.7-Max, Qwen3.8-Flash

PricingUsage from 22 September 2026 is billed at the lower rates.

Source: Together AI, Changelog (opens in a new tab); Together AI, Available models (serverless) (opens in a new tab)

Sep 21, 2026ReleaseGrok 4.7 released on the xAI APIxAI API1 break

xAI released grok-4.7 on a new, larger base model than Grok 4.6, at the same token rates. It has a 500k context window, text and image input and a May 2026 knowledge cutoff.

Platform: xAI API
Model: Grok 4.7

BehaviorOn the Responses API, grok-4.7 always returns reasoning.encrypted_content, even when include does not list it.

Source: xAI, Release Notes (opens in a new tab); xAI, Introducing Grok 4.7 (opens in a new tab); xAI, Grok 4.7 (opens in a new tab)

Sep 18, 2026API changeGemini 2.5 models limited to earlier users on the Gemini APIGemini API1 break

Google limits Gemini 2.5 Pro, Flash and Flash-Lite to users who used them before. The models are not deprecated and keep serving; Google recommends Gemini 3.5 Flash-Lite or Gemini 3.8 Flash for new projects.

Platform: Gemini API
Models: Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-Lite

AvailabilityNew projects cannot start on Gemini 2.5 models.

Source: Google, Gemini API release notes (opens in a new tab); Google, Gemini deprecations (opens in a new tab)

Sep 18, 2026Price changeZ.ai GLM 5.2 cached-input price drops on the Mistral APIMistral API

Mistral's documentation lowered the cached-input price of zai-glm-5-2 in its 18 September 2026 update, to match GLM 5.3. Input and output prices did not change.

Platform: Mistral API
Model: GLM-5.2

Source: Mistral AI, Z.ai GLM 5.2 model card (opens in a new tab); Mistral AI, Z.ai GLM 5.2 model card source, 11 August 2026 revision (opens in a new tab)

Sep 16, 2026ReleaseGemini 3.8 Flash Cyber released for allowlisted customersGoogle Vertex AI

Google released Gemini 3.8 Flash Cyber, a version of Gemini 3.8 Flash for cybersecurity work, behind an allowlist on Gemini Enterprise Agent Platform.

Platform: Google Vertex AI
Model: Gemini 3.8 Flash Cyber

Source: Google Cloud, Model versions and lifecycle (opens in a new tab); Google Cloud, Gemini 3.8 Flash Cyber (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Sep 15, 2026Shutdowngpt-oss-20b and Gemma 4 31B leave Together serverlessTogether AI1 break

Together removed openai/gpt-oss-20b (replacement Qwen3.5 9B) and google/gemma-4-31B-it (replacement GLM-5.3 Flash) from serverless. The removal was announced for 14 September 2026 and is recorded on 15 September 2026.

Platform: Together AI
Models: gpt-oss-20b, Gemma 4 31B
Announced: Sep 1, 2026

AvailabilityRequests to the removed serverless model ID fail. The model can still run on an on-demand dedicated endpoint, billed per minute.

Source: Together AI, Deprecations (opens in a new tab); Together AI, Changelog (opens in a new tab)

Sep 14, 2026ShutdownAmazon Nova Premier reached end of life on Amazon BedrockAmazon Bedrock3 breaks

Bedrock retired Nova Premier on 14 September 2026 in us-east-1, us-east-2 and us-west-2. Requests to the model ID now fail.

Platform: Amazon Bedrock
Model: Amazon Nova Premier
Announced: Mar 13, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.

Source: Amazon Web Services, Amazon Bedrock User Guide (Japanese edition): Model lifecycle (Legacy) (opens in a new tab); Amazon Web Services, Amazon Bedrock GetFoundationModel responses for end-of-life model IDs (opens in a new tab)

Sep 11, 2026ReleaseDeepSeek V4.1 Flash arrives on Together serverlessTogether AI

deepseek-ai/DeepSeek-V4.1-Flash is available on serverless with a context length of one million tokens, function calling and structured outputs.

Platform: Together AI
Model: DeepSeek-V4.1-Flash

Source: Together AI, Changelog (opens in a new tab); Together AI, Available models (serverless) (opens in a new tab)

Sep 10, 2026ShutdownClaude 3 Haiku reached end of life on Amazon BedrockAmazon Bedrock3 breaks

Bedrock retired Claude 3 Haiku on 10 September 2026, including the AWS GovCloud Regions. Requests to the model ID now fail.

Platform: Amazon Bedrock
Model: Claude Haiku 3
Announced: Mar 10, 2026

AvailabilityNew customers can't use Legacy models, and accounts that stop calling one for 15 days may lose access before the end-of-life date.
AvailabilityRequests to the model ID fail after the end-of-life date. Bedrock does not migrate applications to another model.
ToolingYou can't create new Provisioned Throughput or fine-tuning jobs for a Legacy model.

Source: Amazon Web Services, Amazon Bedrock User Guide: Model lifecycle (Legacy) (opens in a new tab); Amazon Web Services, Amazon Bedrock GetFoundationModel responses for end-of-life model IDs (opens in a new tab)

Sep 8, 2026ReleaseGPT-6 Astra available on Amazon BedrockAmazon Bedrock

Bedrock added GPT-6 Astra as openai.gpt-6-astra.

Platform: Amazon Bedrock
Model: GPT-6 Astra

Source: Amazon Web Services, Amazon Bedrock ListFoundationModels, us-east-1 (opens in a new tab)

Sep 3, 2026ReleaseGPT-6 Astra available on Azure AI FoundryAzure AI Foundry

Microsoft made GPT-6 Astra generally available in Foundry with Standard and Provisioned Throughput deployments in Global and US Data Zone geographies.

Platform: Azure AI Foundry
Model: GPT-6 Astra

Source: Microsoft, GPT-6 Astra: Frontier intelligence for work, now generally available in Microsoft Foundry (opens in a new tab)

Sep 3, 2026ReleaseGPT-6 Astra released on the OpenAI APIOpenAI API4 breaks

OpenAI released GPT-6 Astra for reasoning, coding, computer use, research and document creation, in the Responses and Chat Completions APIs.

Platform: OpenAI API
Model: GPT-6 Astra

ParametersGPT-6 Astra does not support the none reasoning effort level.
ParametersGPT-6 Astra does not accept custom temperature or top_p values, or log probabilities.
ToolingTool calling requires the Responses API. Chat Completions requests with tools have to move to Responses.
BehaviorMisalignment monitoring checks agent work in supported Responses API requests and can stop a conversation for review.

Source: OpenAI, Changelog (opens in a new tab)

Sep 2, 2026ReleaseGemini 3.8 Flash releasedGemini API, Google Vertex AI3 breaks

Google released Gemini 3.8 Flash at an introductory price through December 31, 2026.

Platform: Gemini API, Google Vertex AI
Model: Gemini 3.8 Flash

ParametersThe minimal thinking level returns an error. Use low, medium or high.
ParametersReplace thinking_budget with thinking_level, and remove sampling parameters, candidate_count and prefilled model turns.
BehaviorThe model can use more tokens on long, complex tasks by design.

Source: Google, Gemini API release notes (opens in a new tab); Google, What's new in Gemini 3.8 Flash (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Sep 1, 2026ReleaseClaude Fable 5.1 releasedClaude API, Google Vertex AI3 breaks

Claude Fable 5.1 succeeds Claude Fable 5 at the same input and output prices, with cheaper cache reads. Google Cloud lists the same release date.

Platform: Claude API, Google Vertex AI
Model: Claude Fable 5.1

ParametersForced tool choice (tool_choice "any" or "tool") returns a 400 error. Claude Fable 5 accepted it.
BehaviorOnly Claude Fable 5.1 and Claude Mythos 5.1 read Claude Fable 5.1 thinking blocks, and only in the conversation that produced them.
AvailabilityUnlike Claude Opus 5, it requires 30-day data retention and is not available under zero data retention unless Anthropic authorizes it.

Source: Anthropic, Claude Fable 5.1 (opens in a new tab); Anthropic, Claude Fable 5.1 and Claude Mythos 5.1 migration guide (opens in a new tab); Google Cloud, Claude Fable 5.1 on Google Cloud (opens in a new tab); Google Cloud, Gemini Enterprise Agent Platform release notes (opens in a new tab)

Sep 1, 2026ReleaseClaude Mythos 5.1 released for verified organizationsClaude API

Claude Mythos 5.1 offers the capabilities of Claude Fable 5.1 to organizations verified through Anthropic programs.

Platform: Claude API
Model: Claude Mythos 5.1

Source: Anthropic, Claude Mythos 5.1 (opens in a new tab); Anthropic, Claude Fable 5.1 and Claude Mythos 5.1 migration guide (opens in a new tab)