DeepSeek V4.1-Flash-1017 and a search API
deepseek-flash moves to the 1017 version with usage and pricing unchanged. Search and Fetch APIs open at $0.75 per 1,000 calls per endpoint.
AI updates: what changed, and what changes next.
Model releases, price changes, deprecations and shutdowns from the AI vendors teams run in production, each dated and read from the vendor's own page. Last updated 19 October 2026. Follow it by RSS, or read it as JSON or Markdown.
Google Cloud: Gemini 2.5 Pro, Flash and Flash-Lite retire on Vertex AI.
Coming up
32 dated changes ahead, soonest first.
- 20Octtomorrow
Google CloudShutdown
Gemini 2.5 Pro, Flash and Flash-Lite retire on Vertex AI
Vertex AI retires the three Gemini 2.5 models and lists Gemini 3.8 Flash among the replacements. On the Gemini API, 2.5 is limited to existing users, not deprecated.
- 22OctNot before, in 3 days
GoogleShutdown
Veo 3.1 preview ids leave the Gemini API
veo-3.1-generate-preview, -fast- and -lite- stop answering on the Gemini API; Google recommends Gemini Omni Flash. Vertex AI keeps its own dates.
- 23Octin 4 days
OpenAIShutdown
OpenAI shuts down older snapshots and gpt-image-1
Including gpt-3.5-turbo-0125, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, the dated o1, o3-mini and o4-mini snapshots, and gpt-image-1 (replaced by GPT Image 2.5).
- 23Octin 4 days
GoogleShutdown
Deep Research Pro preview shuts down
deep-research-pro-preview-12-2025 stops answering; Google recommends deep-research-preview-04-2026 or its -max- variant.
- 24Octin 5 days
PerplexityShutdown
Older OpenAI model ids leave Perplexity's Agent API
At 00:00 UTC, gpt-5.4, its mini and nano, gpt-5.2, gpt-5.1, gpt-5 and gpt-5-mini retire from the Agent and Orchestrator APIs.
- 31Octin 12 days
Mistral AIShutdown
GLM 5.2 retires on Mistral's platform
Mistral replaces GLM 5.2 with GLM 5.3 on its own platform.
- 31Octin 12 days
OpenAIPlatform
OpenAI Evals become read-only
Existing evals turn read-only ahead of the 30 November shutdown of the Evals platform.
- 31Octin 12 days
DeepSeekPrice
DeepSeek starts billing server-side tool calls
Server-side tool calls through DeepSeek Harness, Claude Code and Codex, free until now, bill from 16:00 UTC.
- 2Novin 14 days
xAIShutdown
grok-imagine-image-quality retires
Requests are served by grok-imagine-image-2.0 at quality low, at a lower per-image price, with the same request and response shape.
- 11Novin 23 days
OpenRouterShutdown
Seedance 1.5 Pro expires on OpenRouter
OpenRouter's own expiration date for bytedance/seedance-1-5-pro. It covers OpenRouter only, not the other providers serving the model.
- 17NovNot before, in 29 days
GoogleShutdown
Gemini preview TTS and Live models shut down
Five preview text-to-speech, Live and native-audio ids leave the Gemini API. Google recommends gemini-3.8-flash-tts, -flash-lite-tts or gemini-3.8-live.
- 19Novin 31 days
Google CloudShutdown
Gemini 3.6 Flash retires on Vertex AI
Vertex AI names Gemini 3.8 Flash as the replacement.
- 19Novin 31 days
Microsoft AzureShutdown
o-series models retire on Azure OpenAI
o1, o1-pro, o3, o3-pro, o3-mini, o4-mini and o3-deep-research retire. Azure recommends GPT-5.6 Sol or Terra.
- 21Novin 33 days
OpenAIPrice
GPT-5.6 Sol promotional price guaranteed until here
OpenAI guarantees the promotional rate at least through this date. Budgets that run past it should use the standard rate.
- 30Novin 42 days
AnthropicShutdown
Claude Sonnet 4.5 retires
Requests to Sonnet 4.5 stop working. Anthropic's replacement is Sonnet 5.5.
- 30Novin 42 days
OpenAIShutdown
Prompts API, Agent Builder and Evals shut down
OpenAI ends the reusable prompts API, Agent Builder and the Evals platform.
- 1Decin 43 days
OpenAIShutdown
gpt-image-1.5 and gpt-image-1-mini shut down
OpenAI recommends GPT Image 2.5 Sunburst or GPT Image 2.5 Flare.
- 1Decin 43 days
GooglePrice
Gemini 4 Argon price doubles
From $2 input and $10 output to $4 and $20 per 1M tokens; output includes thinking tokens.
- 11Decin 53 days
OpenAIShutdown
GPT-5 and o3 dated snapshots shut down
The 2025 GPT-5, GPT-5 mini, nano and pro snapshots and o3-2025-04-16 and o3-pro-2025-06-10 stop answering. OpenAI recommends GPT-5.6 Sol, Terra or Luna.
- 15Decin 57 days
Microsoft AzureShutdown
whisper, tts and tts-hd retire on Azure OpenAI
Azure retires whisper 001, tts 001 and tts-hd 001 and lists no replacement.
- 1Janin 2 months
GooglePrice
Gemini 3.8, 3.7 and 3.6 Flash prices double
From $0.75 input and $3.75 output to $1.50 and $7.50, with context caching doubling the same day.
- 6Janin 3 months
OpenAIShutdown
tts-1 and tts-1-hd shut down
OpenAI recommends gpt-realtime-2.1-mini.
- 14Janin 3 months
OpenAIShutdown
GPT-4.1 and GPT-4o snapshots shut down
gpt-4.1-2025-04-14, gpt-4.1-mini-2025-04-14, gpt-4o-2024-08-06, gpt-4o-2024-11-20 and gpt-4o-mini-2024-07-18 stop answering. OpenAI recommends GPT-6 Luna.
- 20Janin 3 months
OpenAIShutdown
Legacy realtime and audio models shut down
gpt-realtime, gpt-realtime-mini, gpt-audio and the GPT-4o audio models stop answering. OpenAI recommends gpt-realtime-2.1 and gpt-audio-1.5.
- 28Janin 3 months
Google CloudShutdown
Gemini 3.7 Flash retires on Vertex AI
Vertex AI names Gemini 3.8 Flash as the replacement.
- 9Febin 4 months
Microsoft AzureShutdown
GPT-5 2025-08-07 retires on Azure OpenAI
gpt-5, gpt-5-mini and gpt-5-nano version 2025-08-07 retire on Azure; OpenAI's own date for the same snapshots is 11 December 2026.
- 26Febin 4 months
OpenAIShutdown
whisper-1 and GPT-4o transcription shut down
whisper-1, the GPT-4o transcribe models and /v1/audio/translations stop answering. OpenAI recommends gpt-transcribe or gpt-live-transcribe, and realtime translations.
- 15MarNot before, in 5 months
GoogleShutdown
Original Nano Banana leaves the Gemini API
gemini-2.5-flash-image stops answering on the Gemini API; Google recommends Nano Banana 2 Lite.
- 1Aprin 5 months
OpenAIShutdown
GPT-5.4 nano, GPT-5.1 and GPT-5.3-Codex shut down
The models deprecated on 1 October 2026 stop answering.
- 14Aprin 6 months
Microsoft AzureShutdown
GPT-4o and GPT-4.1 retire on Azure OpenAI
The 2025-04-14 GPT-4.1 family and the 2024 GPT-4o and GPT-4o mini versions retire on Azure; OpenAI's API date is 14 January 2027.
- 7MayNot before, in 7 months
GoogleShutdown
Gemini 3.1 Flash-Lite leaves the Gemini API
The earliest shutdown date for gemini-3.1-flash-lite; Google recommends gemini-3.5-flash-lite.
- 2Decin 13 months
European UnionRegulation
EU AI Act rules for high-risk systems apply
The Commission states the rules for high-risk systems — areas such as employment, education and critical infrastructure — apply from this date.
Latest
Newest first. Each entry links the page it was read from.
The last seven days
Prompt-to-voice for GPT-Live
Custom voices from a text description, plus nine new voices. Creating a voice is free; speech bills as GPT-Live sessions.
- gpt-live-1
Claude Haiku 4.5 retirement window opens
Anthropic lists Haiku 4.5 for retirement no sooner than this date.
Fast mode for Opus 5.5 leaves research preview
Fast mode no longer needs an access request or the beta header, and is also available on Microsoft Foundry.
GPT-4.1, GPT-4.1 mini, GPT-4o and GPT-4o mini deprecated
The dated snapshots behind these aliases shut down on 14 January 2027 (gpt-4o-2024-05-13 already on 23 October). OpenAI recommends GPT-6 Luna.
OpenTelemetry export, in beta
Messages API requests and Managed Agents sessions can be traced with OpenTelemetry; the SDKs emit spans in seven languages.
Grok Voice 3.0
Speech-to-speech at $0.08 per minute, the same as 2.0. grok-voice-latest now points to it; grok-voice-2.0 is deprecated with no date set.
- grok-voice-3.0
- grok-voice-2.0
Gemini 4 Argon generally available
Paid tier only, at $2 input and $10 output through 30 November 2026; the rate doubles on 1 December.
- gemini-4-argon
Earlier in October
Model Studio retires legacy Qwen and third-party models
The postponed retirement of older Qwen, Qwen3, audio, DeepSeek, GLM and Kimi models takes effect at 00:00 UTC+8. No replacements are named.
- qwen-turbo
- qwen-vl-max
- qwq-plus
- qwen3-max
- qwen3-coder-plus
- deepseek-v3.2
- deepseek-r1
- glm-4.7
- kimi-k2-thinking
Ultrafast tier for GPT-6.1 Sol
service_tier ultrafast on the Responses API: $12 input, $0.60 cached and $60 output up to 272K input tokens.
Gemini 3.7 and 3.5 Flash deprecated and auto-routed
Requests to gemini-3.7-flash now go to gemini-3.8-flash, and gemini-3.5-flash to gemini-3.6-flash — a different model with no code change.
- Gemini 3.7 Flash
- Gemini 3.8 Flash
- Gemini 3.6 Flash
- gemini-3.5-flash
Claude Haiku 5.5
1M context, 128K output. $0.10 input and $0.50 output up to 100K prompt tokens, $0.50 and $2.50 above. Rejects temperature, top_p, top_k and prefill.
- claude-haiku-5-5
Sonnet 5.5 cache reads cut to $0.10
Prompt cache reads on Claude Sonnet 5.5 drop from $0.20 to $0.10 per 1M tokens. Other rates are unchanged.
Mistral Large 4 in public preview
A multimodal mixture-of-experts model with a 1M context. Mistral says open weights are coming and launch pricing is 50% off for two weeks.
- mistral-large-4
Decisions API in beta
OpenAI's Decisions API runs on GPT-6 Luna. The pricing page lists no Decisions price yet.
Usage tiers cut from five to three
API usage tiers are now Build, Launch and Grow.
Nano Banana 2.1 generally available
$1.50 input, $7.50 text output and $30 image output per 1M tokens. gemini-3.1-flash-image is deprecated, with no date set.
- gemini-nano-banana-2.1
- gemini-3.1-flash-image
GLM 5.3 on Amazon Bedrock
Z.ai's open-weight GLM 5.3 for agentic coding is generally available on Amazon Bedrock.
grok-voice-transcribe-1.0 reaches end of life
Requests to grok-voice-transcribe-1.0 are routed to grok-voice-transcribe-2.0 at the same price.
- grok-voice-transcribe-1.0
- grok-voice-transcribe-2.0
GPT-5.4 nano, GPT-5.1 and GPT-5.3-Codex deprecated
Shutdown on 1 April 2027. OpenAI recommends GPT-6 Luna for GPT-5.4 nano and GPT-6 Sol for GPT-5.3-Codex.
tts-1, tts-1-hd and gpt-4o-mini-tts deprecated
Shutdown on 6 January 2027. OpenAI recommends gpt-realtime-2.1-mini.
September 2026
OCR 4.0 retired after one day's notice
Announced on 29 September for the next day. Mistral recommends OCR 4.1 at the same price.
Claude Sonnet 4.5 retirement announced
Sonnet 4.5 retires on 30 November 2026; Anthropic names Sonnet 5.5 as the replacement.
GPT-6.1 Sol, and an ultrafast tier for Astra
GPT-6.1 Sol at $2 input and $10 output with cached input at $0.10. GPT-6 Astra gains an ultrafast service tier on the Responses API.
Computer use in the Agents API
The Agents API gains computer use.
Claude Sonnet 5.5
$2 input and $10 output, which the launch note calls introductory pricing. Turning off up-front thinking is now thinking type between_tools, not disabled.
Sonar Chat Completions support ends
Synchronous and streaming requests are reformulated as Agent API calls, rolling out by model. Asynchronous Sonar requests must move to the Agent API's background mode.
Sora 2 and the Videos API shut down
OpenAI ended Sora 2 and the Videos API.
Billing resumes for early refusals
Refusals that arrive before any output are billed again in three categories.
Cache diagnostics generally available
Anthropic made its prompt-cache diagnostics generally available on the API.
GPT-6 Sol and GPT-6 Luna
Sol at $2 input, $0.20 cached and $10 output; Luna at $0.10, $0.01 and $0.50.
Claude Opus 5.5
$4 input and $20 output, below Opus 5's $5 and $25. Thinking cannot be disabled: both disabled and enabled thinking settings return a 400.
Gemini 3.8 Flash TTS and Flash-Lite TTS
Both text-to-speech models are generally available, with the new /v1beta/voices endpoint.
- gemini-3.8-flash-tts
- gemini-3.8-flash-lite-tts
Grok 4.7
$2 input, $0.50 cached and $6 output below 200K prompt tokens; $4 and $12 above. 500K-token context, reasoning effort from low to xhigh.
Jev, a System One model
Answers typed questions — yes/no, one of a set, a rubric score — with probabilities instead of text. $0.042 per 1M input tokens; output is free.
Compaction on demand, in beta
A top-level compaction parameter returns a signed compaction block that later requests send in place of the summarized messages.
DeepSeek V4.1-Flash
Served as deepseek-flash, with off-peak prices at half the peak rate. The older deepseek-v4-flash names now route to it.
Agents API in public beta
OpenAI opened the Agents API to everyone as a public beta.
GPT-Live 1 generally available
Voice sessions at $0.05 per minute, billed per second; the backend model and tool usage bill separately.
- gpt-live-1
GPT-6 Astra on Amazon Bedrock
AWS made OpenAI's GPT-6 Astra generally available on Amazon Bedrock.
GPT Image 2.5 Sunburst and Flare
Two image models with new xhigh and max quality settings, billed at GPT Image 2's token rates.
qwen3.8-max now points at the 0902 snapshot
The unpinned qwen3.8-max alias moved to qwen3.8-max-0902 with pricing unchanged. Requests naming the alias got a different model with no code change.
GPT-6 Astra
$10 input and $50 output; above 272K input tokens the whole request bills 2x input and 1.5x output. No custom temperature, top_p or logprobs; tool calling needs the Responses API.
Gemini 3.8 Flash is generally available
1M-token context, 64K output, $0.75 input and $3.75 output until 31 December 2026. Remove temperature, top_p and top_k; thinking_level replaces thinking_budget.
Claude Fable 5.1
$10 input and $50 output, with cache reads cut to $0.25. 1M-token context and 128K output. Forced tool use returns a 400; Anthropic recommends auto with strict tools.
How this timeline is kept
Each entry is read from the vendor's changelog, release notes, pricing, model or deprecation page, which it links with the day it was read. No benchmark scores, no press coverage. Rates are list rates; your provider bills the rate on your own account.
For the story behind September and October, read the roundup; for moving off a retiring model, the deprecation playbook.
When a date on this page is yours
INFRO shows which applications still send a retiring model id, moves traffic to the replacement by weight with a route, and refuses a shut-down model before the provider call with an error that names the replacement.
Questions about AI updates
- What are the latest AI model releases?
- The most recent releases on this timeline: DeepSeek V4.1-Flash-1017 and a search API (DeepSeek, 17 October 2026); Prompt-to-voice for GPT-Live (OpenAI, 16 October 2026); Grok Voice 3.0 (xAI, 14 October 2026); Gemini 4 Argon generally available (Google, 13 October 2026); Claude Haiku 5.5 (Anthropic, 7 October 2026). Each entry links the vendor's release notes.
- Which AI models are being shut down soon?
- Next on the timeline: Gemini 2.5 Pro, Flash and Flash-Lite retire on Vertex AI on 20 October 2026; Veo 3.1 preview ids leave the Gemini API on 22 October 2026 at the earliest; OpenAI shuts down older snapshots and gpt-image-1 on 23 October 2026; Deep Research Pro preview shuts down on 23 October 2026; Older OpenAI model ids leave Perplexity's Agent API on 24 October 2026. A request to a model after its shutdown fails, so find any traffic still sending these ids before the date.
- Where does this information come from?
- From each vendor's own changelog, release notes, pricing, model or deprecation page — never from press coverage. Every event names its page and the day INFRO read it. A date a vendor gives as "no earlier than" is marked as such.
- Are the prices here what I will pay?
- They are the vendors' published list rates, per million tokens unless the entry says otherwise. Your provider bills you at the rate on your own account, so a negotiated rate replaces them. INFRO adds 0% to model spend; it charges a subscription for the control plane.
- How do I follow new entries?
- Subscribe to the RSS feed at https://infro.io/ai-updates/feed.xml, or read the same list as JSON at https://infro.io/ai-updates.json or Markdown at https://infro.io/md/ai-updates.md.