Google introduced Gemini 4 Argon, a reasoning model for long software, enterprise and cyber tasks. It launches first to trusted defenders, then paid API and AI Ultra users, with lower intro pricing and a 1M-token output limit.

OpenAI’s Dots are always-on GPT-6 Astra agents that handle ongoing work across apps and devices, retain context, learn user preferences, and operate under permissions, review controls, and enterprise governance.

OpenAI launched GPT‑6.1 Sol, offering near‑Astra results for coding, computer use and business workflows at about one‑fifth the cost, with cached input at $0.10 per million tokens and stronger accuracy, safety and tool use.

Meta launched Muse for Small Business, an AI agent that connects Meta tools and third-party apps to support planning and operations. It can use business data across services, with approval required before posting, messaging or spending.

Anthropic launched Claude Sonnet 5.5 as a faster, lower-cost upgrade to Sonnet 5 for everyday work. It delivers over 30% faster output, lower task costs, strong benchmark gains, broader platform support, and new cyber safeguards.

Manus 2.0 expands across web, desktop, and mobile with a new agent system, Studio, automations, game and video tools, device control, and Cue personal agents, while reporting lower token use, faster task completion, and reduced cost.

ElevenLabs launched Eleven v4 for high-quality scripted speech and v4 Turbo for real-time voice agents, combining multi-speaker control, audio tags, 90+ languages, and low-latency streaming via its products, API, and agents.

Leaks suggest Anthropic may launch Claude Sonnet 5.5 as early as next week. Reports point to stronger performance, up to 1M context, and competitive pricing as Anthropic, OpenAI, and Google race to release new flagship AI models.

Meta is preparing two Muse web additions: a persistent screen-view tab for monitoring tasks and browser-based voice calling. The updates expand Muse beyond chat, though access remains limited to supported markets.

Google has relabeled references to Nano Banana 2.5 Flash as Nano Banana 2.1 in Flow, suggesting a smaller, iterative image-model update. No release date or feature details are confirmed, and gains remain unclear.

OpenAI appears to be preparing a broader Ultrafast API rollout near DevDay, with new doc references and a hidden speed selector. The mode may expand beyond limited access, with cost, model support, and compute tiers in focus.

OpenAI may unveil “O,” an always-on agent, at DevDay on September 29. References in ChatGPT config and the Pro plan suggest a persistent assistant with possible email support, though features, permissions, and access remain unclear.

Microsoft is recasting Copilot around Home, Code and Autopilot, uniting chat, delegated work, app creation and persistent agents in one app, with staged previews, Office integration, enterprise controls and metered pricing.

Google launched Gemini 3.8 Live Avatar for enterprise, adding near real-time video personas to conversational AI. It supports tool calls, multilingual speech across 97 languages, custom avatars, and SynthID watermarking.

OpenAI may launch a $500/month ChatGPT Pro Max plan aimed at developers and researchers, centered on fastest Work and Codex access. It could add higher limits and longer task capacity, with possible ties to Cerebras-powered inference.

Meta unveiled 100g VR Glasses with 5K micro-OLED displays, a separate compute puck, gesture and voice controls, IMAX support, immersive sports, private work screens, and hologram calls, priced at $1,299.99 for spring 2027.

Mitra’s new AI assistant manages calls, texts, email, scheduling, web tasks, and follow-ups from a user’s own accounts. It hands off routine work, lets users monitor or intervene live, and runs on a subscription with no data sales.

OpenAI appears to be preparing API onboarding with Free, Prototype, and Accelerate plans, guiding developers from testing ideas to production. The structure suggests a broader platform vision, though app hosting remains unconfirmed.

Leaked Gemini 4 Pro tests suggest Google may have fixed its frontend design weaknesses. Early outputs show polished web and SVG work, longer generation times, and signs of a true frontier model ahead of rivals like Anthropic.

OpenAI is poised for a Tuesday launch after Tibo Sottiaux confirmed the date, fueling speculation that GPT-6 Sol, Terra, and Luna may arrive before DevDay. Users also hope the update addresses tight usage limits.

Meta’s Muse is gaining traction fast, outpacing ChatGPT’s early mobile launch in the US and Canada. Meta is now preparing Muse for Messenger and Telegram, pointing to a broader cross-platform agent strategy for users and businesses.

Ahead of Meta Connect on September 23, Meta’s developer platform shows signs of wider rollout, including Video Playground, Voice Playground, Connectors, and possible Oracle Cloud ties, pointing to broader API and enterprise plans.

Xiaomi open-sourced MiMo-V2.6 Pro and Flash, two omnimodal models for coding, vision and computer use. Pro leads open-source rankings, Flash cuts cost, and UltraSpeed delivers up to 20x faster output for low-latency tasks.

SpaceXAI launched Grok 4.7 across coding tools, APIs, and cloud platforms, offering stronger coding, longer task focus, lower-cost pricing, and tougher safeguards, with benchmark gains over Grok 4.6 and free testing in Grok Build.

Anthropic appears to be quietly A/B testing Claude Fable 5.2, with users reporting stronger one-shot JavaScript outputs and richer canvas demos. Early comparisons suggest a clear step up from 5.1, as Opus 5.5 rumors also build.

SpaceXAI launched Grok Voice Transcribe 2.0, claiming 2x the accuracy of v1.0 at unchanged pricing. It supports multilingual, real-time and batch transcription, with diarization, timestamps, term biasing, and phone-audio gains.

Meta is testing a Mail tab for Muse that may give the agent a dedicated mailbox or centralize messages from connected inboxes. The feature aligns with Shared Agents and Muse’s broader task automation, but details and timing remain unconfirmed.

xAI added memory to Grok Build, letting it retain team conventions, decisions, project facts, and commands across sessions. Notes are stored by topic and scope, reviewed after each turn, and applied in later work.

Anthropic is unifying Claude chat and Cowork into one workflow, letting users switch from quick questions to delegated work in one thread. Docs, Slides, and Design join conversations in beta for paid plans, with rollout starting on Pro and Max.

TypeSafe launched Jev, a typed decision model for software that returns bounded, schema-safe outputs with confidence scores. It targets routing, scoring, and extraction with low latency and cost, though benchmark claims warrant caution.

Mistral and Mozilla are bringing private, multilingual AI to Firefox Smart Window, with regional language tuning, zero data retention, and support for search, tab discovery, and recalled content, starting in France and North America.

Leaked screenshots point to Google testing a math-focused DeepThink V3 variant, “Mathematica,” tuned for heavy calculation and symbolic reasoning, with a 1M-token context. Public release remains unconfirmed pending internal tests.

Apodex 1.1 shifts from answering questions to completing analytical work across files, with code execution, traceable outputs, mid-task recovery, and multi-agent coordination, posting large benchmark gains over 1.0.

Google is updating Gemini Notebook with real-time voice chat, mobile recording, interactive study reports, new quiz types, and short video overviews, while expanding free higher-limit plans for eligible college students.

Meta One rolls out globally with paid tiers across Instagram, Facebook, WhatsApp and Meta AI. It adds 50+ features for consumers, creators and businesses, while core apps and standard Meta AI access remain free.

Google launched Gemini 3.8 Live and Extended Thinking, voice models for near real-time dialogue, reasoning and task execution. Live serves high-volume, multilingual work; Extended Thinking handles complex workflows and leads key speech benchmarks.

NoimosAI is adding customer messaging to its autonomous marketing platform, with an agent that builds lifecycle email flows from plain-language prompts, drafts branded messages, supports review before sending, and tracks product goals.

ElevenLabs added ElevenCreative to its hosted MCP server, letting Claude, ChatGPT, Cursor and coding agents create audio, images and video in-chat via one OAuth sign-in, alongside existing agent management tools.

Google appears close to launching Interactive Reports in Gemini Notebook: longer reports with embedded mind maps, slides, quizzes, and flashcards generated on demand, combining research, teaching, and explainer formats in one document.

Anthropic is testing a Claude “Money” tab in its mobile app that would let users link bank accounts and ask spending and budgeting questions. The unreleased feature points to a possible US-first launch, though timing remains unclear.

Cursor Projects, now in beta, coordinates long-running software work through delegated agents. It runs tasks in cloud or local environments, supports scheduled and PR-driven maintenance, and targets work spanning multiple PRs.

OpenAI launched ChatGPT for Financial Services, combining financial data and GPT-6 Astra for research, modeling, and client materials, with enterprise controls, source citations, and integrations shaped by Morgan Stanley and Evercore.

OpenAI released GPT-Live-1 in the API, a full-duplex voice model for apps and workflows that listens and speaks at once, supports interruptions, cuts latency, and offers prompt-based control, telephony use, and separate backend pricing.

Meta launched Muse in the US on iOS, Android, web, and WhatsApp. Testing points to unreleased features including voice mode, custom voice creation, hidden reasoning controls, and a possible Muse Spark model update.

Meta’s Muse app contains an unannounced Shared Agents section, suggesting users may soon build, configure, and share specialized sub-agents. The move aligns with Meta’s cross-app AI strategy and could target businesses and creators.

Inception launched Mercury 2.5, a diffusion LLM claiming 40% higher intelligence than Mercury 2, 1,107 tokens per second, a 260K context window, and low pricing, targeting fast search, voice, and coding workloads.

OpenAI’s ChatGPT Images 2.5 brings faster image generation, sharper output, tighter edit control, and stronger subject consistency. It adds Sketch, templates, and API models Flare and Sunburst for consumer and production use.

Meta unveiled Muse, a persistent AI agent for long-running personal tasks. It works in the background, uses web and system tools, keeps context across chats, and adds logs, approvals and memory controls for oversight and safety.

Google is internally testing a Projects feature for Gemini desktop: persistent workspaces with custom instructions, files, and grouped chats. It mirrors long-standing ChatGPT and Claude setups, but no public launch details are confirmed.

Keygraph released Shannon 3.0, an open-source AI pentester for web apps and APIs. It analyses code, validates findings only through live exploits, adds CI/CD support, resumable scans, and reports with evidence under AGPL-3.0.