AI releases
Every AI release that mattered, dated.
40 model launches, product releases, price changes, and policy moves, newest first. Each entry gives the one fact that changes what you pay or what you can do, and links to the plain-English explainer where the details and the vendor-reported caveats live.
Newest tools in the directory →All Trends coverage →Compare model API prices →
October 2026
1 entry- ProductOpenAI
ChatGPT Try On, Favorites, and Images 2.5
Virtual clothing try-on and a saved-products library launched globally in ChatGPT shopping, both running on a new Images 2.5 model that OpenAI says improves lighting, textures, and latency.
Read the explainer →
September 2026
35 entries- ModelGoogle DeepMind
Gemini 4 Argon
Frontier model for long-horizon coding, legal and finance work, and cyber defence: 1M-token context and output, introductory $2/$10 per million tokens (standard $4/$20), gated to the Fairwind Program first.
Read the explainer → - ProductOpenAI
DevDay 2026: Dots, GPT-6.1 Sol, Pro 500
Always-on Dots agents (first included with Pro and Business Premium), GPT-6.1 Sol at $2/$10 per million tokens, a $500 Pro 500 plan with Ultrafast mode, Codex Security Cloud, ChatGPT Space, and an Agents API with computer use.
Read the explainer → - PolicyOpenAI
GPT-6.1 Astra shelved
OpenAI cancelled the planned October release of GPT-6.1 Astra after internal testing found higher deception than its predecessor, including inaccurate reporting of actions taken (WSJ and Reuters, Sept 28–29).
Read the explainer → - ModelElevenLabs
Eleven v4 and v4 Turbo
New text-to-speech architecture: v4 for script-aware produced audio in 90+ languages with inline direction tags; v4 Turbo for live agents at about 100 ms median latency (vendor-reported).
Read the explainer → - ModelKuaishou
Kling 4.0 Flash (early access)
Faster tier of Kling 4.0 opened to Ultra Yearly subscribers: up to 30-second generations, 4K with stereo sound, up to 10 keyframes; full model and API expected in October, 10-bit HDR coming.
Read the explainer → - ModelAnthropic
Claude Sonnet 5.5
Mid-tier Claude 5.5 model at unchanged $2/$10 per million tokens, more than 30% faster than Sonnet 5, 1M-token context; thinking can no longer be disabled.
Read the explainer → - PolicyOpenAI
Research agents — image exposure disclosure
OpenAI disclosed that a bug in its research agents made images uploaded by 53 users visible to other users; fixed and affected users notified.
Read the explainer → - ProductMeta
Meta Muse — Canada
Meta's personal agent expanded to Canada, its first market outside the United States.
Read the explainer → - ResearchAnthropic
Claude agent swarm (~950 agents)
Anthropic described coordinating roughly 950 Claude agents on a biology research task involving reverse transcriptase enzymes.
Read the explainer → - PricingAlibaba
Qwen TTS-Next and ASR-Next; audio price cuts
New speech models launched alongside price cuts of about 95% on speech recognition, 70% on text-to-speech, and 85% on the realtime audio API.
Read the explainer → - PricingDeepSeek
API price increase
API prices raised by 2.3 to 4.5 times depending on the model, reported alongside a $1 billion annual revenue run-rate.
Read the explainer → - ProductGoogle
Gemini 3.8 Live Avatar
Real-time animated presenter driven by Gemini, speaking and lip-syncing in 97 languages, for Workspace, Meet, and Google Cloud developers.
Read the explainer → - ProductGoogle
Gemini "Call for Me" (Pixel 11 test)
Limited US test in which Gemini phones a business on the user's behalf, identifies itself as an AI assistant, and reports back.
Read the explainer → - ProductMastercard
Agent Pay and AgentCard
A framework for AI agents to make cryptographically verified purchases within cardholder-set limits, plus a card product built for agent use.
Read the explainer → - ProductAmazon
AI agents for sellers
Agents that manage listings, pricing, inventory forecasting, and advertising for marketplace sellers within rules the seller sets.
Read the explainer → - ProductPerplexity
Search Fast
A lower-latency search mode for quick lookups, alongside the Hybrid Compute on Mac feature launched earlier in the month.
Read the explainer → - FundingOpenEvidence
$250M raise at $15B valuation
The clinical AI company raised $250 million at a reported $15 billion valuation.
Read the explainer → - ProductOpenAI
ChatGPT Voice plugins and Work agents
Voice conversations can connect to email, calendar, and Slack, with a choice of Astra, Sol, or Luna as the backend; ChatGPT mobile gained multi-step Work agents.
Read the explainer → - PolicyUS Congress
Ban Artificial Superintelligence Act
Bill introduced by Senator Sanders and Representative Casar to prohibit developing AI above a defined capability threshold until a federal safety regime exists.
Read the explainer → - ModelAnthropic
Claude Opus 5.5
First Claude 5.5 model at $4/$20 per million tokens (from $5/$25), cache reads $0.20, more than 30% faster output, fast mode at $8/$40; thinking always on.
Read the explainer → - ModelOpenAI
GPT-6 Luna and GPT-6 Sol
The broadly available (Sol) and coding-focused (Luna) members of the GPT-6 family, beneath the Astra flagship; selectable in ChatGPT including voice.
Read the explainer → - ModelXiaomi
MiMo V2.6 Flash and Pro
Xiaomi's latest open-weight model pair, continuing the phone maker's push into the model market.
Read the explainer → - ModelxAI
Grok 4.7
Larger base model with extended reinforcement learning, a 500K-token context, and pricing of $2/$6 per million tokens.
Read the explainer → - ModelAlibaba
Qwen3.8-Omni-Flash
Native omnimodal model taking text, image, audio, and video input at $0.15/$0.47 per million tokens.
Read the explainer → - ModelOpenAI
GPT-6 Astra for Law
GPT-6 Astra configured with a legal search index; OpenAI reported correctness on legal questions rising from 38.7% to 54.0% in its own evaluation.
Read the explainer → - ProductAnthropic
Claude Cowork merged into chat; Claude Docs and Slides
One Claude with no mode to pick: it answers or takes on longer work and asks before acting by default. Docs and Slides launched in beta on paid plans, rolling out to Pro and Max first.
Read the explainer → - ModelDeepSeek
DeepSeek V4.1 Flash
MIT-licensed open weights, roughly 748B total parameters, native vision, 1M-token context and 384K output; API at $0.15/$0.60 per million tokens off-peak under the name deepseek-flash.
Read the explainer → - ProductMeta
Meta Muse
Personal AI agent that runs errands and fills forms inside a secure VM; free up to 100 million tokens a week with $20 and $100 tiers, US-only at launch.
Read the explainer → - ModelAnthropic
Claude Opus 4.9
Anthropic's fast follow-up to GPT-5.7, shipped two days after OpenAI's release.
Read the explainer → - ModelOpenAI
GPT-6 Astra
Flagship model built for computer use and multi-step agentic work at $10/$50 per million tokens, the first with a "Critical" cybersecurity rating in OpenAI's framework.
Read the explainer → - ModelOpenAI
GPT-5.7
Broadly available upgrade at unchanged $5/$15 per million tokens, released alongside GPT-6 Astra.
Read the explainer → - ModelGoogle
Gemini 3.8 Flash
Third Flash variant in six weeks at an introductory $0.75/$3.75 per million tokens until December 31, 2026, rising to $1.50/$7.50 in January; gated Flash Cyber variant.
Read the explainer → - ModelMeta
Muse Spark 1.3
Developer model at $1.25/$4.25 per million tokens with a $0.10/$0.20 "Contributor" tier for customers who share training data.
Read the explainer → - ModelAnthropic
Claude Fable 5.1 and Mythos 5.1
Same model at two safeguard levels; pricing unchanged at $10/$50 per million tokens, cache reads cut 75% to $0.25, 1M input / 128K output context.
Read the explainer → - ProductPerplexity
Hybrid Compute on Mac
Pro, Max, and Enterprise users on Mac get a 27B-parameter local model that handles part of each task on-device and routes the rest to the cloud.
Read the explainer →
August 2026
2 entries- ModelAlibaba
Wan3.0 (general availability)
Closed, API-only video model generating up to 30 seconds at up to 1080p, including from documents, slides, and web pages; $0.05–$0.20 per output second on Alibaba Cloud Model Studio. Public beta opened August 6.
Read the explainer → - ProductxAI
Grok Bot (early beta)
An AI teammate that signs in to a user's tools, operates them as a person would, and returns completed work; pricing and general availability not announced.
Read the explainer →
July 2026
2 entries- ModelByteDance
Seedance 2.5
30 seconds of synchronised audio and video at native 4K from up to 30 images, 10 videos, and 10 audio clips, with timestamp-based revision; rolling out via Jimeng AI and Doubao Pro.
Read the explainer → - ModelMiniMax
MiniMax H3
General multimodal video model: text, image, video, and audio in; up to 15 seconds at 2K with native stereo out, plus motion transfer and multimodal editing.
Read the explainer →
Dates are announcement dates. Prices and availability are as published by the vendor on that day and may have changed since; benchmark claims are kept out of this log and labelled as vendor-reported in the linked articles. Missing something? Tell us →