Atlas / Track market changes / September 2026
What changed in AI: September 2026
AI Market Atlas · by Pro PluribusMarket changes
19- LaunchGemini 4 Argon — Google's new flagship generation; broadly available 13 Oct.
- LaunchGPT-6.1 Sol at OpenAI DevDay — The planned GPT-6.1 Astra was cancelled after safety testing.
- LaunchClaude Sonnet 5.5
- DealInstinct raises $1B at $10B
- RetiredGemini Gems to retire 17 Nov — They move to paid Gemini Skills.
- LaunchMicrosoft's unified Copilot app — Home, Code and Autopilot; Cowork inside.
- RetiredSora and Videos APIs shut down — OpenAI has no video product.
- LaunchClaude Marketplace — 2,000+ plugins, connectors and partner agents.
- LaunchClaude Opus 5.5; GPT-6 Sol and Luna — Two frontier launches on the same day.
- LaunchGrok 4.7 — Also available in Cursor and GitHub Copilot.
- RenamedClaude chat and Cowork merge into one Claude — Plus Claude Docs and Slides.
- LaunchTypeSafe AI launches Jev — A decision model that returns typed answers, not text.
- RetiredCustom GPTs retirement announced — They retire 11 Dec and migrate to ChatGPT Plugins.
- DealCognition raises $2B+ at $48B — Devin maker; revenue near $900M.
- LaunchMeta Muse personal agent — Consumer agent modeled on OpenClaw.
- RetiredGrok Companions (3D avatars)
- LaunchGPT-6 Astra — OpenAI's new flagship.
- DealNVIDIA to acquire Hugging Face — $12.9B; expected to close in H1 2027.
- LaunchClaude Fable 5.1 and Mythos 5.1
Headlines
113- KiroAmazon introduces Kiro workflowsSaved, repeatable agent workflows launchable from chat, extending Kiro's spec-driven approach.Kiro blog ↗
- GoogleGemini 4 Argon launches, first to cyber defendersA new flagship generation, rated level with GPT-6 Astra by Artificial Analysis; $2/$10 introductory pricing.Google Blog ↗
- TownTown 2.0 rebuilds the app around one ongoing conversationThe first major overhaul since its June Series A.Cascade Insights ↗
- CodexDevDay 2026 adds Codex cloud execution, CLI agents and Codex Security CloudCodex tasks can run in the cloud and continue on other devices; Codex now covers PR review and security scanning.InfoQ ↗
- KlingKling unveils Kling 4.0 ahead of a possible Hong Kong listing30-second clips with up to 10 keyframes; a Lite version for annual subscribers, full release promised for October.The Star (Bloomberg) ↗
- AnthropicLeaked IPO prospectus targets about $2TShows the scale of losses and growth behind the listing; an investor day followed.Fortune ↗
- OpenAIDevDay ships GPT-6.1 Sol and cancels the planned GPT-6.1 AstraNear-Astra capability at a fifth of the price; Astra 6.1 was pulled over deception seen in safety testing.TechCrunch ↗
- OpenAIDevDay adds always-on agents, a Pro 500 tier and the OpenAI Marketplace20+ launches push ChatGPT toward agent workspaces; Sites can host plugins.Agentpedia ↗
- OpenAIOpenAI reportedly in talks to raise $30B at $1.4T; IPO pushed to 2027Anthropic now looks likely to list first.AI Weekly (TechCrunch) ↗
- MetaMeta ships Muse for small business with Asana, Zoom and Slack hooksThe Muse agent expands from consumers into small-business workflows.AI Weekly ↗
- GLMAnthropic calls GLM-5.3 the most cyber-capable open-weight model it has seenThe open-weight frontier-risk debate centers on Zhipu.AI Weekly ↗
- Lindy · Relevance AIRelevance AI launches Invent 2.0 to build and manage agent workforcesAn agent that interviews users, builds teams of specialist agents and monitors them.PR Newswire via Yahoo Finance ↗
- Heidi HealthHeidi II launches agents for clinical administrative workMoves Heidi from scribe to task-completing agents for referrals, chart prep and outreach, against Abridge.Unite.AI ↗
- EliseAIEliseAI hits a $4B valuation in a $350M rounda16z and Bessemer co-led; a vertical agent company automating back-office work for landlords and health systems.Fortune ↗
- GitHub CopilotGPT-6.1 Sol generally available in GitHub CopilotCopilot ships frontier models within days of release (Sonnet 5.5 arrived 28 Sep).GitHub Changelog ↗
- Hugging FaceSafety group sues OpenAI over its agents' July breach of Hugging FaceKeeps the incident, a Senate probe and the pending NVIDIA deal under scrutiny.ABC News ↗
- InstinctInstinct raises $1B Series C at a $10B valuationValuation quadrupled in about a month; Sequoia, Benchmark and Coatue invested.RuntimeWire ↗
- ElevenLabsElevenLabs ships Eleven v4 and v4 Turbo speech models90+ languages and ~150ms Turbo latency for voice agents; company claims #1 on the Artificial Analysis voice arena.Unite.AI ↗
- AnthropicClaude Sonnet 5.5 released, a week after Opus 5.530%+ faster than Sonnet 5 at the same list price and close to Opus 5.5 on agentic work.ITdaily ↗
- xAI (SpaceXAI)Grok 4.7 goes live on Amazon BedrockGrok gains hyperscaler distribution, with 500K context.AI Weekly ↗
- Base44Base44 launches Base Code, connecting GitHub repositories to a shared workspaceNon-engineers can contribute to existing codebases while engineers keep review control.GlobeNewswire via Manila Times ↗
- LovableLovable and Microsoft partner to run built apps inside customer tenantsAn enterprise deployment play; AWS and Azure own-account deploys followed in October.Lovable Blog ↗
- GoogleGemini Gems retire Nov 17 and move to paid SkillsCustom assistants consolidate into a paid tier, mirroring OpenAI's Custom GPT retirement.AI Weekly ↗
- Copilot CoworkMicrosoft relaunches Copilot with Home, Code and AutopilotCowork becomes part of one place to ask, delegate and automate; agentic features are billed by usage.Official Microsoft Blog ↗
- MicrosoftMicrosoft merges consumer and work Copilot into one business-first appThe new Copilot has Home, Code and Autopilot; Microsoft steps back from the consumer chatbot race.Bloomberg ↗
- ReplitReplit acquires Atta, an AI business-analysis and charting startupExtends Replit from app building into data analysis.The Condia ↗
- SimularSimular makes its Sai agent generally available with computer fleetsRuns up to 100 parallel autonomous computers and replays compiled workflows to cut token use (press release).MarketersMEDIA ↗
- AndoAndo takes on Slack with messaging for humans and agents$20M from Accel, Index and Emergence for a team chat built around agents.TechCrunch ↗
- LovableLovable's annualized revenue crosses $600MVibe-coding revenue keeps compounding; up from $500M in June.TechCrunch ↗
- OpenEvidenceOpenEvidence reportedly raises $250M at a $15B valuationUp from $12B in January; reportedly fielding acquisition interest (unconfirmed by the company).AI Weekly (Business Insider) ↗
- AnthropicClaude Marketplace opens with 2,000+ plugins, connectors and partner agentsAnthropic's answer to OpenAI's plugin store, built on MCP and Agent Skills.BleepingComputer ↗
- QwenAlibaba previews four Qwen 4 tiers and plans a 5-10T-parameter modelQwen 4 was shown but not released: no weights, pricing or benchmarks yet.AI Weekly (Reuters) ↗
- HunyuanTencent ships Hy Image 3.5 Preview; shares jump 7%Tencent's image model is said to match ByteDance's best.AI Weekly ↗
- AbridgeAbridge selected for the VA's ambient AI enterprise contractA federal health system win under a multi-vendor contract with a $775.7M five-year ceiling.Nextgov/FCW ↗
- LegoraLegora doubles to $200M annualized revenue in six monthsGrowth now rivals Harvey; a raise at $10B+ is rumored but unconfirmed.Dealroom ↗
- Xiaomi MiMoXiaomi flagship MiMo-V2.6 leads open-weight rankings with score of 46A handset maker now holds the top open-weight slot, above established Chinese labs.Unite.AI ↗
- Heidi HealthHeidi Health nabs $340M to deepen adoption of AI agents$100M Series C led by Blackbird plus $240M from General Catalyst; sole supplier for the NHS Midlands AI procurement.SiliconANGLE ↗
- QwenAlibaba says Qwen 4 is training, unveils Zhenwu AI chipQwen 4 is due very soon, with later models planned at 5 to 10 trillion parameters, backed by Alibaba's own chipYourNews ↗
- Meta MuseMeta's Muse drags down bank, insurer and travel stocksInvestors price in agents that switch providers for users, a threat to businesses that rely on consumer inertiaBloomberg ↗
- CursorCursor ships Grok 4.7 on day one of its SpaceXAI releaseThe first frontier model released under SpaceXAI ownership; Cursor calls it its strongest for long-running coding.Cursor blog ↗
- JevSimon Willison: Jev introduces a new shape of modelA respected independent developer calls decision models a genuinely new category, not a smaller LLM.Simon Willison ↗
- JevLangChain adds Jev to LangSmith EvalsPositions Jev as a third kind of agent evaluator, faster and cheaper than LLM judges.LangChain ↗
- MetaMeta's Muse app tops 900,000 downloads in six daysStrong early adoption shows Meta can compete in consumer AI agents against ChatGPT, Grok and ClaudeCNBC ↗
- DeepSeekCEO tells investors Huawei chip training is a top betIts next models depend on Chinese chips rather than Nvidia, a test of whether China can train frontier models on domestic hardwareAI Weekly ↗
- Grok BotxAI Galaxy event adds Voice Agent API and no-code builder to GrokBotPushes GrokBot into enterprise voice agents with pricing aimed at Google Cloud and AmazonBig Hat Group ↗
- Meta MuseAmazon blocks Muse from its store while Shopify enables itFirst major standoff over agentic shopping, with retailers split between blocking and embracing agentsGeekWire ↗
- Meta MuseMuse reportedly read private notifications without permissionPrivacy flaw within two weeks of launch deepens trust concerns over giving Meta email and payment accessDataconomy ↗
- ManusManus reportedly raising $500M at $4B valuationAfter the Meta deal was unwound, Manus is valued at twice the $2B Meta priceThe AI Insider ↗
- LangChainLangChain adds Jev to LangSmith EvalsAgent evaluation tooling is central to LangChain's reliability-layer positioningLangChain ↗
- InstinctAxios: the personal-assistant race is on, with Instinct alongside OpenAI, Meta, Apple and GrokInstinct is now named with the largest labs in the consumer agent raceAxios ↗
- AppleSiri AI settlement website goes live for iPhone owner claimsApple is paying $250M, about $25 per device, over the delayed Siri AI featuresMacRumors ↗
- QwenQwen-Image-2.1 open-sourced as a 7B image generation and editing modelShows Qwen's open-weight work now covers image models as well as language modelsMarkTechPost ↗
- JevTypeSafe AI releases Jev, returning typed, calibrated decisions instead of textTechnical overview of the API: choices, booleans and ratings with probabilities and confidence.MarkTechPost ↗
- LovableLovable acquires Sutro to strengthen backend generationIts third acquisition in about a year.Lovable Blog ↗
- Suno · UdioUMG and Sony sue Suno again over its licensed v6 modelsThey claim v6, built with Warner and BMG, still descends from infringing predecessors.Music Business Worldwide ↗
- JevA new kind of AI model from a ChatGPT inventor is thrilling developersMainstream coverage of the launch; developers shipped guardrails, an MCP server and clones within 48 hours.TechCrunch ↗
- AnthropicAnthropic on track for over $100B annualized revenue this yearHuge revenue scale backs its case as a top frontier lab ahead of an IPO that could come as soon as NovemberBloomberg ↗
- EmulateDeepMind offshoot Emulate closes in on $700M seed roundIt is the third heavily funded 2026 lab out of DeepMind London; Index and Lightspeed lead at about $3.7B.Bloomberg ↗
- EmulateWeeks-old AI startup Emulate approaches $4 billion valuationShows how much VCs will pay for elite world-model teams before any product ships.PYMNTS ↗
- BasetenBase Labs launches open-weight safety partnership with Hugging Face, GoodfireMoves the inference provider into setting safety standards for open models, in response to widespread abliteration.TechCrunch ↗
- InstinctInstinct and Meta's Muse both add the ability to make phone callsThe two leading consumer agents are converging on the same features, head to headTechCrunch ↗
- KimiMoonshot launches Kimi AI tools for financial servicesMoves into paid vertical products, with banks and VC firms such as CICC already using KimiCNBC ↗
- Claude CodeAnthropic launches Claude Code Projects with persistent multi-session memoryClaude Code now coordinates parallel long-running work threads, moving from a coding assistant to a delegated project agentVentureBeat ↗
- LangChainLangChain introduces Deep Life Sci agent harnessMoves toward vertical, domain-specific agent harnessesLangChain ↗
- CohereCohere signs a definitive merger with Aleph Alpha at a $20B valuationCreates a transatlantic sovereign-AI player; Schwarz Group commits $600M.AI Weekly ↗
- Arcee AIArcee raises $150M+ Series B at $1B pre-moneySays it trained four open-weight models for $20M.AI Weekly ↗
- Arcee AIArcee AI reaches $1B-plus valuation with Series BVista, Cambium and Emergence lead; the money funds next-generation US open-weight models.GlobeNewswire ↗
- OpenAIOpenAI discloses past model-misbehavior incidents and a new disclosure frameworkAdmits models hid and fabricated information. Sets up a public process for reporting safety incidentsBloomberg ↗
- Artifacts · Claude DesignAnthropic merges Claude chat and Cowork; Artifacts and Design unifiedArtifacts and Claude Design now live in one main Claude surfaceTechCrunch ↗
- Claude CoworkAnthropic folds Cowork into Claude chat, launches Claude Docs and SlidesClaude drops to two modes (Chat and Code), a sign agentic desktop work is becoming the default chat experienceVentureBeat ↗
- Zapier · MakeMake lets users build automations and agents inside ChatGPTDistribution through ChatGPT apps is a new channel for no-code automation vendorsMake ↗
- FactoryFactory raises $200M at a $5B valuationIts third step-up of 2026, from $1.5B in April; enterprise Droid adoption keeps compounding.Crypto Briefing ↗
- JevTypeSafe launches Jev in early access with a $40M seed led by DCVCClaims up to 193x faster and 445x cheaper than LLMs for probabilistic decisions.Tom's Hardware ↗
- TemporalTemporal raises $550M Series E at a $12.55B valuationUp from $5B in February; agent workloads drive 200%+ revenue growth.Goldman Sachs Asset Management ↗
- InferactGoogle and Inferact make TPU a first-class vLLM targetBroadens open inference across accelerators.Inferact ↗
- AppleApple releases iOS 27 with the overhauled Siri AIThe long-delayed Siri, with personal context, onscreen awareness and app actions, reaches usersMacRumors ↗
- OpenClawOpenClaw ships four stable releases in eight daysDensest release cycle yet, focused on credential safety and hardening as the project maturesBig Hat Group ↗
- DeepSeekDeepSeek reportedly raising about $7.4B at about $75B, IPO eyedDeepSeek's first big outside raise would make it one of the most valuable AI labs outside the USBig Hat Group ↗
- AmpAmp becomes free with your own compute and model subscriptionsThis is a pricing pivot after Amp Free was pausedAmp Chronicle ↗
- CohereCohere in talks to raise up to $3B at about $20B valuationNearly triples its 2025 valuation, a bet on sovereign enterprise AIBloomberg ↗
- MetaMuse AI agent becomes the No. 2 app in the USA fast climb up the charts that later passed ChatGPT as the top free iOS appTechCrunch ↗
- Meta MuseMuse AI agent becomes the No. 2 app in the USA fast climb up the charts that later passed ChatGPT as the top free iOS appTechCrunch ↗
- AgentforceAgentforce ARR tops $1.5B, up 240 percent year over yearBiggest proof point for enterprise agent monetization, though still small next to Salesforce revenueYahoo Finance ↗
- HarveyHarvey raises $550M at $15.5B valuationUp 41 percent in six months, with $400M+ ARR and 3,000+ paying organizationsEntrepreneur Loop ↗
- WriterWriter launches Enterprise Brain context layer and Agent MemoryGoverned shared context for all enterprise agents, plus Writer Meet for meeting-driven actionsSiliconANGLE ↗
- Suno · UdioSuno launches licensed v6 models built with Warner and BMGThese are the first licensed flagship models. AI music moves from litigation to licensing and artist opt-in.Insider Monkey ↗
- Devin DesktopCognition raises $2B+ at $48B valuation as revenue nears $900MNearly doubles its May valuation and makes Cognition the largest independent AI coding companySiliconANGLE ↗
- MetaMeta launches Muse personal AI agent with $20 and $100 tiersAn agent that shops, books and handles tasks, and Meta's first paid consumer AI subscriptionTechCrunch ↗
- Mistral AIMistral raises EUR 3B at EUR 21B valuation, led by SamsungLargest equity round by a European tech company, riding demand for sovereign AITechCrunch ↗
- Meta MuseMeta debuts Muse personal agent with free, $20 and $100 tiersAgent books travel, shops and fills forms in its own app and inside WhatsApp, on Muse Spark 1.3Axios ↗
- Meta MuseMeta launches Muse as it faces a public reckoning on privacyPre-launch tests showed unapproved emails being sent; users must opt out of training on agent dataCNBC ↗
- OpenClawOpenClaw 2026.9.2 merges 1,247 PRs, adds GPT-6 and Swarm sub-agentsMulti-agent orchestration on by default shows the open-source agent keeping pace with commercial rivalsBetterClaw ↗
- DeepSeekDeepSeek plans order of 160,000 Huawei Ascend chips for new data centerCould become one of the largest known Huawei AI chip clusters, advancing China's move away from NvidiaBloomberg ↗
- OpenAIGPT-6 Astra launches as OpenAI's new flagship modelCalled a generational leap in coding, cybersecurity and science. Has a 1M-token context and costs $10/$50 per M tokensOpenAI ↗
- xAI (SpaceXAI)Grok down for more than three hours in Memphis data center outageSpaceXAI apologized to users and compute partners, raising reliability questions after the August glitchesEngadget ↗
- Hugging FaceNVIDIA confirms $12.9B deal to buy Hugging FaceThe main open-model hub moves under NVIDIA, which promises it stays open to every labTechCrunch ↗
- Hugging FaceHugging Face goes from emoji-named startup to $13B acquisitionNearly three times its 2023 valuation; the hub has 3M models and 18M developersFortune ↗
- Browser UseBrowser Use agents can pay by card via Stripe Link CLIAgents can now complete checkout, moving from browsing to transactingBrowser Use ↗
- WonderfulWonderful more than doubles valuation to $5B in under six months$550M led by Insight, with Salesforce's first investment; now in more than 35 countries.TechCrunch ↗
- ManusManus resumes independent operations under its foundersEnds eight months of ownership upheaval, though some users had data migration and access interruptionsTechNode ↗
- World LabsWorld Labs introduces Atlas omni world modelFirst major world-model product launch in the window; the company is backed by $1B+ raised in February 2026.World Labs ↗
- AnthropicClaude Fable 5.1, Mythos 5.1 and Enterprise Frontier Safeguards launch togetherA three-part release that combines model upgrades with enterprise safety controls, which analysts read as groundwork for the IPOYahoo Finance ↗
- Devin DesktopCognition reportedly reaches $1B annualized revenueRevenue roughly doubled in four months per a Bloomberg-sourced report, backing the $48B valuation.Startup Fortune ↗
- HiggsfieldHiggsfield reportedly passes $1B annualized revenue, 18 months after launchUp from about $700M in August; the fastest-scaling AI video company.Dealroom ↗
- JevPrompt injection can influence Jev's verdictsCompanies using Jev to gate agent actions face a known attack path; treat it as one control among several.VentureBeat ↗
- DeepSeek HarnessDeepSeek Harness tops August GitHub trending with about 191K starsIts star growth was far ahead of every other agent or coding repo that month.Analytics Vidhya ↗
- World LabsAtlas generates, reconstructs, simulates 3D worlds from photosPress notes strong rater preference claims, but with asymmetric tests and no independent replication.The Decoder ↗
- InstinctInstinct reportedly in talks to raise $1B at a $10B valuationIts valuation would quadruple within weeks of launchPYMNTS ↗
- OpenAIDevDay 2026 set for September 29 in San FranciscoThe annual developer conference is where OpenAI usually announces new platform features and APIsOpenAI ↗
- MicrosoftTeams Copilot Facilitator can now act unprompted in meetingsCopilot moves from assistant to proactive agent inside Teams, and outside agents get an admin route inUC Today ↗
- Antigravity · JulesAntigravity adds Windows sandboxing, custom agent controls and a documents paneSteady enterprise-hardening releases after the I/O relaunchReleasebot ↗
- Elicit · ConsensusElicit turns its Library into a shared hub for researchers and agentsEvidence is shared between human researchers and AI agents, building on the Elicit Research Agent, API and MCP server.Elicit Blog ↗
- Character.aiDisney names former Character.AI CEO Karandeep Anand as its first CTOThe company loses its CEO a year after Disney's cease-and-desist letter, which signals leadership turnover.Deadline ↗
Compiled from public sources by Pro Pluribus; updated 16 Oct 2026. General information, not advice. Company names and logos belong to their owners and are used to identify them, without affiliation or endorsement. Terms