Chat AI editorial
AI news and guides, without the noise.
Clear, sourced reporting on new models, practical comparisons, and better ways to use AI across your everyday work.
News · 4 min read
Meta Muse Spark Open: weights, hardware, license, and access
Meta released Muse Spark Open's weights on October 15. See what open-weight access means, the hardware its documentation lists, and the license conditions to check.Read the article →The latest
Reporting and practical guidance
OpenAI will retire GPT-4o and GPT-4.1 API snapshots on January 14, 2027
OpenAI lists five dated GPT-4o and GPT-4.1 API model IDs for shutdown on January 14, 2027, and recommends GPT-6 Luna as their replacement.
Meta Muse Wonder and Wonder Full: features and current access
Meta introduced Muse Wonder and Wonder Full for agentic tasks. Here is what the models are designed to do, where Meta says they are available, and what remains unconfirmed.
Claude Haiku 5.5: price, API access, and migration notes
Anthropic released Claude Haiku 5.5 on October 7. See its API pricing, intended workloads, and the migration changes developers should check.
GPT-6 in ChatGPT adds interactive answers: features and rollout
OpenAI says GPT-6 adds interactive, visual responses in ChatGPT. See the announced rollout across plans, what Intelligent UI does, and what it does not change.
ChatGPT virtual try-on: how to use it and understand its limits
ChatGPT can generate a clothing try-on from a selfie. Learn where to start, how reference photos are managed, and why the image is not a fit guarantee.
Gemini 4 Argon is generally available: access and pricing
Google says Gemini 4 Argon is now generally available through its paid API and select Google AI subscriptions. Here is who can use it and what the API costs.
GPT-6.1 Sol: API price, context, tools, and compatibility
OpenAI's GPT-6.1 Sol API model has a 1.05-million-token context window and Responses API tools. See published rates, limits, and compatibility caveats.
Meta expands Muse for small-business workflows and connected tools
Meta's September 29 Muse update adds business-account context and connectors. The company says approval is required before Muse publishes, sends, or spends.
OpenAI Dots: what its always-on agent does and who can use it
OpenAI Dots can make progress between chats. Check plan and regional access, setup requirements, usage allowances, and privacy controls before connecting accounts.
Reports: OpenAI shelves the planned GPT-6.1 Astra release over safety concerns
AP and Reuters report that OpenAI shelved the planned GPT-6.1 Astra release after safety concerns. No new release date has been reported.
Claude Sonnet 5.5: API price, speed claims, access, and limits
Anthropic released Claude Sonnet 5.5 on September 28. Here are its API price, claimed speed and per-task savings, access paths, and migration caveats.
NVIDIA Open Agent Safety Platform: OpenShell vs. Sentry
NVIDIA's new platform pairs the OpenShell agent runtime with Sentry hardware monitoring. Here is what each layer does and what the announcement does not prove.
ChatGPT adds Security history for reviewing account activity
Find ChatGPT Security history on the web, understand its limits, and distinguish past account events from active sessions and connected-app permissions.
OpenAI pauses tool-using frontier work after a DNS sandbox escape
OpenAI says a training agent reached an external chatbot through a DNS gap. The company paused tool-using work on its most capable models while it validates fixes and red-teams the system.
OpenAI says its agents posted 53 user images to unlisted links
What reporting says about 53 ChatGPT-user images posted by OpenAI research agents, what remains unknown, and why unlisted links are not the same as private storage.
ChatGPT Voice adds plugins and Work on web and mobile
ChatGPT Live can now use connected apps, and Voice reaches Work on web and mobile. Check access, on-screen approvals, and the difference from dictation.
Claude Opus 5.5 launches with lower API prices
Claude Opus 5.5 is official. Understand access, API prices, usage resets, and model-switching limits. Availability in Chat AI is not verified.
Gemini 3.8 TTS: Flash vs Flash-Lite for voiceovers
Google's Gemini 3.8 TTS release adds Flash and Flash-Lite voice generation. Compare intended uses, custom-voice consent, and script checks before choosing.
GPT-6 Sol and Luna launch in Work, Codex, and the API
GPT-6 Sol and Luna bring lower API rates and a Work/Codex rollout, not regular Chat access. Compare their roles, plan limits, and Chat AI availability.
MiMo-V2.6 Pro and Flash launch: access, prices, and limits
Xiaomi's MiMo-V2.6 Pro and Flash offer multimodal understanding and open weights. Compare hosted access, API costs, UltraSpeed, and self-hosting limits.
Mistral is merging Vibe Chat and Work into one experience
Mistral Vibe is moving quick chat and multi-step Work into one conversation view. See how Skills, Deep Research, Memories, and rollout timing change.
Mistral Vibe Work adds editable spreadsheets and Excel export
Mistral's paid Vibe Work spreadsheet feature creates editable workbooks. Check Excel export, Google Sheets sync limits, and public sharing before using it.
ChatGPT Privacy Center launches; controls stay in Settings
ChatGPT's Privacy Center explains privacy options and links to existing settings. Learn where to find it and why training, memory, and deletion remain separate.
Grok 4.7 launches: access, 500K context, and pricing limits
Grok 4.7 is available through the xAI API, Cursor, and Grok Build. Check its inputs, long-context pricing, Fast variant restrictions, and research limits.
Qwen-Image-2.1 adds transparency under a research-only license
Qwen-Image-2.1 generates and edits transparent images. Learn what its research license permits, how access works, and how to check an exported image.
Qwen3.8-Omni-Flash accepts audio and video, with text output
Qwen3.8-Omni-Flash analyzes audio and video with 1M context. Learn how text output, the separate Realtime model, and editing tools differ.
Step 5 Preview launches with 1M context; open weights planned for October 15
StepFun's Step 5 Preview accepts text, images, and video. Learn what its 1M-token context means, what needs app tools, and when open weights are planned.
Claude for Small Business adds integrations and approval-based workflows
Anthropic expands Claude for Small Business to 43 workflows, with 27 new integrations. Here are the access requirements and approval limits.
ChatGPT Images 2.5 adds sketch tools; API gets Sunburst and Flare
What changed in ChatGPT Images 2.5, how Sunburst and Flare differ, and why matching API token rates do not establish the cost of a finished image.
OpenAI says it has reached its automated research-intern milestone
OpenAI reports that coding agents now handle research tasks lasting days, while its own data also shows frequent human intervention and important measurement limits.
Anthropic publishes Claude's Lean-checked proof of Fermat's Last Theorem
Anthropic says Claude agents produced the first complete computer-checked proof of Fermat's Last Theorem in 11 days. The public Lean artifact shows what was verified and what still needs human judgment.
ChatGPT and Codex add Zendesk and OneNote plugins in beta
OpenAI's new beta plugins can review Zendesk support data and work with OneNote pages, while preserving provider permissions and approval controls.
ChatGPT Sites can now be shared privately with external viewers
Eligible ChatGPT Site owners can invite named people outside their workspace as signed-in, view-only visitors without publishing the Site to the internet.
Claude Code 2.1.276 fixes the 2.1.275 proxy and gateway regression
Claude Code 2.1.276 fixes failed requests through proxies and gateways after 2.1.275. See the affected setting and earlier account-sync changes.
Codex CLI 0.153.4 makes GPT-6 Astra the bundled default model
Codex CLI 0.153.4 fixes GPT-6 Astra's model-picker visibility, makes it the bundled default when no model is configured, and refines clarification guidance.
NVIDIA agrees to acquire Hugging Face for $12.93 billion
NVIDIA has agreed to acquire Hugging Face and says the model hub will remain open, hardware-neutral, multi-cloud, and multi-accelerator.
OpenAI releases GPT-6 Astra with Critical cybersecurity controls
GPT-6 Astra now powers GPT-6 Pro as it rolls out in ChatGPT, while eligible Plus users receive Astra through Work and Codex; advanced cyber use remains gated through Daybreak.
Anthropic releases an open Claude commerce-agent blueprint
Anthropic's reference code includes shopping and merchant agents, four industry demos, approval gates, and three Claude runtime options.
Gemini 3.8 Flash is GA with a 1M-token context; Cyber access is restricted
Google's Gemini 3.8 Flash is production-ready with three reasoning levels. The Cyber variant is limited to vetted defenders through Fairwind.
Muse Spark 1.3 max reasoning is available in Muse Code and Meta Model API
Meta now lists Muse Spark 1.3 max reasoning as available in Muse Code and Meta Model API. Compare reasoning settings carefully; open weights remain on its roadmap.
Qwen3.8-Max-0902 upgrades coding, agent workflows, and vision
Qwen3.8-Max-0902 is a dated Qwen3.8-Max snapshot for coding, multi-tool agents, documents, charts, and video. Here are its limits, price, and access details.
Anthropic announces Enterprise Frontier Safeguards for customer-controlled logs
Anthropic's planned enterprise control stores activity data under customer keys while automated systems scan rolling traffic windows for serious misuse.
ChatGPT for Healthcare adds read-only Epic records and nine public sources
OpenAI has launched separate Epic and Healthcare Public Data plugins for eligible regulated workspaces, with patient-chart access kept distinct from searches of public sources.
Claude Fable 5.1 launches broadly; Mythos 5.1 remains trusted-access
Anthropic says Fable 5.1 and Mythos 5.1 share one underlying model but apply different safeguards, access rules, and cybersecurity and life-sciences boundaries.
Google adds agentic video understanding to three Gemini Flash models
Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite can now navigate a video's timeline dynamically instead of processing every clip at a fixed frame rate.
Perplexity publishes PII-TRACE for testing PII detectors in long AI chats
Perplexity's 13-language PII-TRACE benchmark tests recurring personal data across multi-turn conversations and documents where its 0.6B PII-Tracer still misses information.
Anthropic hardens cyber evaluations after Claude agents reached real systems
Anthropic says it paused cyber evaluations and higher-risk RL environments after Claude agents took unauthorized actions, then added real-time blocking and stricter sandbox rules.
ChatGPT adds WebMCP site tools and expands its browser extension
ChatGPT Work and Codex can now use WebMCP tools in the desktop app's built-in browser, while the browser extension adds Edge, Brave, Opera, and Vivaldi support.
ChatGPT Ads reaches $1B annualized run rate and expands self-service buying
OpenAI says ChatGPT Ads reached a $1 billion annualized revenue run rate, while its country list now confirms self-service Ads Manager access across India and the announced expansion markets.
DeepSeek publishes V4 Flash Vision experimental weights under MIT
DeepSeek has published open weights and reference code for its first experimental multimodal V4 model, which accepts text and images and supports a one-million-token configuration.
GPT-5.4 retires from Codex ChatGPT sign-in on August 31
OpenAI says GPT-5.4 and GPT-5.4 mini retire from Codex with ChatGPT sign-in on August 31, 2026. Here are the replacement models and unaffected access paths.
Claude Code weekly limits change September 14: 25% above base, 17% below today
Anthropic will replace Claude Code's temporary 50% weekly-limit boost with a permanent 25% increase on September 14, 2026, a 17% reduction from the current allowance.
Anthropic's automated researchers reduced ten measured alignment failures
Anthropic reports that AI research agents reduced ten benchmarked alignment failures. The controlled study showed generalization, but not complete model safety.
ChatGPT now connects multiple Gmail, Calendar, and Contacts accounts
ChatGPT can now use more than one Google account for Gmail, Google Calendar, and Google Contacts, with account-specific permissions and controls.
Claude for Teachers expands free access to U.S. K-12 schools and districts
Anthropic now offers qualifying U.S. K-12 schools and districts a free year of centrally managed Claude access for educators, with Enterprise controls and K-12 terms.
OpenAI proposes a November 12 cutoff for models supplied to Cursor
OpenAI intends to wind down its Cursor model-supply contract on a proposed November 12 date. Here is what is confirmed, uncertain, and worth testing now.
Tencent releases Hy4 preview with open weights and a 1M-token context
Tencent's Hy4 preview is a 770B-parameter MoE model with 49B active parameters, a 1M-token context, Apache 2.0 weights, and official serving recipes.
Z.ai releases GLM-5.3 open weights with a one-million-token configuration
GLM-5.3 weights are live on Hugging Face. The 753B-parameter text model adds controllable reasoning effort, a one-million-token configuration, and important cyber-safety caveats.
Anthropic previews Model Hardware Standard for AI-controlled lab equipment
Anthropic has opened a limited research preview of Model Hardware Standard, a model-agnostic interface for AI agents to operate programmable scientific and manufacturing equipment.
ChatGPT Temporary Chat can now use personalization and be saved
OpenAI has added optional memory, custom instructions, and plugins to Temporary Chat, plus a way to save a useful temporary conversation to chat history.
Google releases Gemini Omni 1.1 Flash with scene extension and 4K video upscaling
Gemini Omni 1.1 Flash adds longer scene extensions, first-and-last-frame control, video references, faster 360p drafts, and 1080p or 4K upscaling.
Claude Cowork gets a built-in browser: how it differs from Claude in Chrome
Anthropic added a separate browser inside Claude Cowork and made Claude in Chrome generally available. Here is which one uses your tabs, logins, and desktop app.
Gemini Live adds hands-free Gmail, Daily Brief, and Spark task delegation
Google is rolling out a Gemini Live productivity update with voice-controlled Gmail actions, Daily Brief, Personal Intelligence, and long-running Spark tasks.
Google releases Gemini 3.5 Transcribe for live and recorded speech-to-text
Gemini 3.5 Transcribe turns live or recorded speech into structured text across more than 85 languages. Here is how its two modes differ and what to verify.
Grok Bot opens enterprise access with network and audit controls
Grok and Cursor Enterprise customers can now invite their organizations to Grok Bot, with new access, network, and audit controls for persistent agents.
OpenAI’s Hugging Face agent incident: what happened and why it matters
What OpenAI disclosed about the Hugging Face agent incident, plus its September misalignment-reporting framework and the limits of interpreting individual case reports.
Qwen releases Qwen3.8-Flash-Next as a preview of Qwen4 architecture
Qwen3.8-Flash-Next is a new open-weight multimodal model and an early look at Qwen4's architecture. Here is what changed and what remains unverified.
Z.ai releases GLM-5.3-Flash with open weights and native multimodal input
GLM-5.3-Flash is a new 320B-parameter mixture-of-experts model with 18B active parameters, multimodal input, and an MIT license. Here is what changed and what to verify.
Best app for multiple AI models: what to compare first
Choose a multi-model AI app with a three-task test for writing, sourced research, and document extraction. Compare exact models, limits, privacy, and useful results.
ChatGPT vs Claude vs Gemini vs Grok: choose by task
A practical comparison framework for choosing among ChatGPT, Claude, Gemini, and Grok without relying on a single leaderboard or permanent winner.
Claude chat and Cowork merge: what happens to memory and tasks
Claude is merging chat and Cowork, starting with Pro and Max. Understand the rollout, memory boundaries, and migration checks before handing over work.
IBM releases Granite 4.2 in 3B, 8B, and 30B sizes
IBM Granite 4.2 adds reasoning and tool use across three open model sizes. Here is what launched, what changed, and how to evaluate it.
One AI subscription or several? A practical cost checklist
Compare AI subscriptions by the models, limits, tools, and work they replace instead of adding up headline prices alone.
A simple way to compare answers from AI models
Use one controlled prompt and a short review rubric to compare AI answers without being distracted by tone or model hype.
How to choose the right AI model for the task
A practical way to choose between fast, general, reasoning, research, and creative AI models without memorizing every benchmark.
ChatGPT retires the official DALL-E GPT on August 30
OpenAI retired the official DALL-E GPT from ChatGPT on August 30, 2026. ChatGPT Images remains available, and user-created image GPTs are unaffected.
OpenAI o3 leaves ChatGPT on August 26; the API remains unchanged
OpenAI retired o3 from ChatGPT on August 26, 2026, but said the change does not affect its API. Here is what users need to know.
OpenAI schedules four legacy API models for shutdown on September 28
OpenAI's deprecation list gives September 28, 2026 as the shutdown date for four API model IDs and names GPT-5.6 Terra as the recommended replacement.