Replit opens a waitlist for its Windows desktop app preview, which builds and runs apps locally in sandboxes powered by Microsoft Execution Containers and NVIDIA OpenShell.
Perplexity Decider v1.1 arrives on OpenRouter at $0.02 per million input tokens with free output, returning structured answers and probabilities from text or images.
Microsoft Execution Containers reach general availability on Windows 11; GitHub Copilot’s local-model routing enters experimental preview later in October.
RTX Spark laptops open for preorder with up to 128GB unified memory for local AI; shipping starts October 16, with compact desktops due in November.
VS Code 1.141 adds agent sandboxing on Windows, macOS and Linux, plus handoff of local Codex chats from ChatGPT or the CLI with conversation history intact.
Tibo: A banked usage reset is coming to paid Codex and ChatGPT Work accounts by the end of October 7, Pacific time.
GPT-6 with Intelligent UI starts rolling out to ChatGPT Plus, Pro, Business and Enterprise on October 7; Free and Go follow October 8, with Enterprise access subject to admin settings.
Claude Max and Team API credits roll out this week: $100 monthly for Max 5x, $200 for Max 20x, and up to $500 pooled for Team; new subscribers must wait seven days.
Claude Sonnet 5.5 cache reads now cost $0.10 per million tokens, down 50%.
Claude Haiku 5.5 launches at $0.10 input and $0.50 output per million tokens for prompts up to 100K; longer prompts cost $0.50 input and $2.50 output.
SynthID Detector opens globally in English, checking images, video and audio for AI watermarks from Google, OpenAI, NVIDIA and Kakao; Apple support is coming.
Isomorphic Labs joins the Virtual Biology Initiative to build an open biological-data resource for AI models that predict and help treat disease.
Muse’s iPad app is now available.
Henri B. Kagan and Kenso Soai win the 2026 Nobel Prize in Chemistry for discovering non-linear effects and autocatalysis in asymmetric organic synthesis.
ChatGPT Work’s new-thread bug has a fix for macOS and Linux; Windows users are still waiting for an updated desktop app.
Grok Bot will use outside models, including Claude Opus 5.5, Midjourney and Suno, selecting a backend for each task.
Claude’s usage dashboard and Admin API usage reports have recovered from errors; Messages API requests were unaffected.
OpenAI’s Workspace Agents have recovered after an incident that could prevent users from seeing responses.
Vercel AI Gateway adds a beta that reruns uncertain decisions with a fallback model when confidence drops below your chosen threshold.
Cursor’s iPhone app now lets you start tasks, reply to and monitor agents running on your computer; Enterprise access requires admin approval.
OpenAI releases 722 AI-generated mathematics manuscripts and supporting proofs; verification is incomplete, and some results may contain errors.
Nano Banana 2.1 is generally available for image generation and editing; gemini-3.1-flash-image is deprecated, with no shutdown date announced.
Google releases EmbeddingGemma 2, a 740-million-parameter open model for searching text, code, images, video and audio directly on devices.
ChatGPT’s Meetings plugin launches in beta for Pro and Business users on Mac, saving meeting summaries and action items in ChatGPT Space.
OpenAI’s Decisions API opens in public beta, letting apps choose models, tools and actions; OpenAI claims decisions are up to 10× faster than GPT-6 Luna via Responses.
OpenAI replaces five API usage tiers with three: Build, Launch and Grow, with automatic upgrades based on total credit purchases.
Anthropic expands its Cyber Verification Program to three access tiers, giving approved security professionals access to models including Claude Mythos 5.1.
GitHub stacked pull requests are now available on all github.com plans: split large changes into smaller reviews, then merge them together.
Claude Code 2.1.292 closes sandbox file-access gaps and adds controls for how much effort subagents spend on tasks.
Crew-12 targets ISS undocking no earlier than October 7 at 12:05 UTC, followed by splashdown off California on October 8 at 15:34 UTC, weather permitting.
Only 4.5% of consumers in YipitData’s U.S. panel paid for ChatGPT, Gemini or Claude in August, up from 2.1% a year earlier, according to a16z.
Google DeepMind’s AlphaProtein Novo designs enzymes to make piperidine and break down the pollutant DEHP, according to a new preprint; code is available.
Gemini 4 Argon ranks first in Arena’s text and creative-writing tables; scores are preliminary, and Google’s initial rollout is limited to trusted cyber defenders.
Francis Halzen wins the 2026 Nobel Prize in Physics for his work on IceCube and the discovery of high-energy cosmic neutrinos.
ChatGPT conversations, image generation, Spaces, Pages, API rehydration and Dots have fully recovered from the latest elevated-error incident.
Codex Auto-review is now free for users signed in with ChatGPT and no longer draws from plan usage; enable it under Settings → General → Permissions.
Claude Code 2.1.291 fixes lost permission replies in cloud sessions and missing final messages when quitting.
ChatGPT conversations, image generation, Spaces, Pages, API rehydration and Dots hit elevated errors; mitigation is applied and recovery is being monitored.
Falcon 9’s fourth SDA Tranche 1 data transport mission now targets October 8, following the aborted launch attempt.
iPhone Duo app submissions are open ahead of its October 23 launch; Apple requires Duo screenshots for submissions starting April 2027.
GitHub secret scanning now detects Lovable API keys, Pydantic credentials and additional Supabase tokens.
Google Docs adds native Markdown editing and collaboration; Drive adds formatted previews, rolling out from October 5 over up to 15 days.
Claude Code 2.1.290 fixes missed or duplicate scheduled tasks and closes permission gaps in plan mode and linked instruction files.
Codex CLI 0.160.1 fixes Windows remote MCP startup when custom environment variables are configured.
Googlebook goes on sale with Gemini, Android phone integration and 10 years of automatic updates.
Codex Cloud and ChatGPT conversation errors are mitigated and under monitoring; OpenAI’s Admin Console remains intermittently inaccessible.
ChatGPT and Codex will watermark eligible EU text in coming weeks; API users worldwide can opt in on select models now, with detection initially restricted to approved researchers.
Tibo: GPT-6 Astra and GPT-6.1 Sol get roughly 50% faster default speeds across subscriptions and Sign in with ChatGPT partners, with no user changes needed.
Meta AI adds parental supervision globally: view teens’ recent topics, not conversations, and select stricter content limits.
Claude Mythos 5.1 and Fable 5.1 recovered after errors from 12:40–13:10 UTC on October 5.
15 years after Steve Jobs’s death, Tim Cook: “His spirit lives on”; Apple CEO John Ternus: “Forever grateful.”
MiniMax’s LLM service had repeated 1–5-minute error incidents on October 5; its latest incident was resolved at 11:04 UTC.
ChatGPT will test visual ads during image generation in the US later in October; ads stay separate from the generated image.
ChatGPT Work and Codex in ChatGPT Desktop errors are mitigated; recovery remains under monitoring.
Qwen Code 0.25.0 adds collaborating workspace agents, agent-to-agent sharing and IMAP/SMTP email support.
Falcon 9 aborts the SDA Tranche 1 launch just before liftoff; the rocket and payload are healthy, with another attempt no earlier than October 6.
Google expands free AI training to 650,000 educators in the UK and Ireland.
2026 Nobel Prize in Physiology or Medicine recognizes Karl Deisseroth, Peter Hegemann, and Georg Nagel for their pioneering work on channelrhodopsin and optogenetics.
Codex/Work gets a 28-day pledge from Tibo: a meaningful improvement or a full reset each day.
Hark targets a launch this week, promising free paid-plan access to its first 100,000 sign-ups.
Vercel’s experimental fx 0.0.13 adds OpenAI Ultrafast support through AI Gateway, adjustable auto-compaction and shell-approval fixes.
GPT-Rosalind billing is scheduled to start October 5: $5 per million input tokens and $25 per million output tokens for approved life-sciences research.
Google’s September AI push spans nine billion DNA-change predictions, hourly weather forecasts and roughly 300 languages.
Grok 4.7 in Cursor passes 23/23 tasks and leads VulcanBench’s Frontier v4 at extra-high effort; different AI judges limit direct comparisons.
Tibo: GPT-6.1 Sol Ultrafast coming soon.
OpenClaw’s Android update clears review after Pichai offered to follow up.
Vercel reports a KVM security flaw; details pending.
Elon Musk backed Trump's rename of AI as SI.
Musk posts “We’re so BACK!!” after praise for Grok Bot’s speed.
Meta open-sources Muse’s ESP32/Linux SDKs and announces Home Link.
VISTA paper reports Opus 5.0 cleared all 25 public ARC-AGI-3 games.
Eric Schmidt says AI now writes code he cannot.
Google’s May Antigravity preview is scheduled to retire October 5.
Claude Code 2.1.289 fixes permission bypasses.
GitHub Copilot code reviews now support REST and GraphQL.
GitHub Copilot retires Gemini 3.5/3.6 Flash, Kimi K2.7 Code and Claude Opus 4.7.
Grok transcription 1.0 now routes to 2.0 at the same price.
Codex CLI 0.160.0 fixes duplicate sends and Windows sandbox issues.
GitHub Copilot previews desktop-app control on macOS and Windows.
Google’s Data Agent Kit for coding agents reaches general availability.
Latest Live news