122 public catches this week · all weeks · the live wire

07-19RTNhacker news

An HN story suggests Claude Code is now using Bun, which is written in Rust.

07-19URGENTbinary

A new model ID, "claude-sonnet-4-5fLF", was detected in the @anthropic-ai/claude-code-darwin-arm64 native binary, alongside new endpoints and internal feature flag strings.

07-19RTNgithub_releases+pages

Claude's code verification and code review skills now require explicit invocation with /verify or /code-review, rather than running automatically.

07-19RTNbinary

Anthropic's Claude Code binary for macOS ARM64 (2.1.214) shows new internal development strings, including mentions of 'internal error' conditions, 'empty_tree_fallback_enabled', and several `getInternalField` and `createInternalModuleById` calls, alongside GitHub PR linking and notification preference API endpoints.

07-18URGENTct_logs

Three new subdomains, including 'stt-flux-multi-chs2.titanium.api.anthropic.com', appeared in Anthropic's certificate transparency logs, suggesting potential new API or staging infrastructure related to speech-to-text.

07-18RTNsitemap

Anthropic quietly published a new documentation page for migrating to managed agents.

07-18URGENTgithub

Anthropic's Python SDK v0.117.0 adds support for "dreaming" and "MCP Tunnels" features.

07-18URGENTgithub_releases+pages

Claude Code adds an EndConversation tool for abusive users, a progress heartbeat for long tool calls, and various telemetry attributes.

07-18FLASHpypi

Anthropic's latest SDK (0.117.0) hints at two new Claude model IDs, `claude-opus-4-8` and `claude-haiku-4-5`, along with new beta features for 'dreams' and 'tunnels' via specific API endpoints.

07-18URGENTnpm

Anthropic's SDK update includes new 'tunnels' endpoints, a 'myContext' request option, and adds various JSON examples for tool use and conversational prompts.

07-18RTNweb app

Anthropic's Claude platform console deployed new frontend code with 16 new API endpoints, many related to workspace and API key management within organizations.

07-18FLASHlmarena

LMArena has added 5 new models to its leaderboard, including two anonymous models (kivine-wxzc, saga-decima-0kzi), and three named/preview models (claude-fable-5-v2-image-to-webdev, claude-sonnet-5-image-to-webdev, gemini-3.5-flash). The anonymous models are likely stealth models being tested.

07-16URGENTgithub_org

Anthropic released a new public GitHub repository: "code-migration-kit-with-claude-code," which includes prompts, templates, and scripts for large-scale language migrations using Claude Code.

07-16RTNbinary

Anthropic's `claude-code` binary adds new strings related to background tasks, quiet hours, and references to `code.claude.com` documentation and a `/v1/design/grants` endpoint.

07-15RTNgithub_releases+pages

Claude Code released v2.1.211 with a new flag for subagent text in stream-json output, plus numerous bug fixes and performance improvements.

07-15FLASHhacker news

A Hacker News post titled "I tricked Claude into leaking your deepest, darkest secrets" is gaining rapid velocity, suggesting a potential new vulnerability or exploit for Anthropic's Claude model.

07-15RTNct_logs

A new subdomain, identity.anthropic.com, appeared in certificate transparency logs.

07-15URGENTbinary

Anthropic's native Claude Code binary for macOS ARM64 (2.1.210) adds strings related to internal package/snippet hosting, secrets management, and new memory endpoints. ✓ beat the announcement by 20.1h

07-15RTNgithub_releases+pages

Anthropic's claude-code released v2.1.210 with numerous bug fixes, improved tool messaging, and enhanced auto mode permissions defaulting to Sonnet 5 for external sessions.

07-14URGENTdocs

Anthropic updated their API changelog to announce a new Admin API for Claude Enterprise, allowing user and organization management, available in beta.

07-14RTNweb app

Anthropic's Claude platform console shows a new internal_tier feature flag and endpoint, possibly for organization-specific internal access or tiers.

07-14URGENTsitemap

Anthropic's sitemap shows 17 new pages related to API admin functions, user management, and multiagent orchestration, suggesting upcoming enterprise-focused RBAC and agent capabilities.

07-14RTNweb app

Anthropic's Claude platform updated its frontend code with new API endpoints related to billing and payment methods.

07-14RTNsitemap

Anthropic added new sitemap URLs related to K12 legal documents and 'Claude for Teachers', suggesting a potential focus on educational use cases.

07-14URGENTgithub_org

Anthropic released a new public GitHub repo named 'k12-teacher-skills' focused on AI skills and evaluation rubrics for K-12 teachers, co-developed with Learning Commons. ✓ beat the announcement by 0.0h

07-14RTNhacker news

An HN post gaining velocity discusses strategies to prevent Claude from using the phrase 'load-bearing'.

07-14URGENTbinary

Anthropic's Claude code for macOS ARM64 updated with new internal feature flags and potential API endpoints, suggesting development on background tasks, error handling, and OAuth settings. ✓ beat the announcement by 37.8h

07-14RTNgithub_releases+pages

Anthropic's claude-code released v2.1.209, fixing a bug that blocked dialogs in `claude agents` background sessions.

07-14RTNgithub_releases+pages

Claude Code v2.1.208 introduces a screen reader mode, vim insert mode remaps, and a process wrapper, alongside numerous bug fixes.

07-13RTNhacker news

An HN post likening Claude to Mr. Meeseeks is gaining traction, suggesting it's good for single, simple tasks.

07-13URGENTdocs

Anthropic's API changelog indicates 'Dreams' (research preview) now supports Claude Fable 5 and Claude Sonnet 5. ✓ beat the announcement by 22.9h

07-13RTNweb app

Anthropic's Claude console has new API endpoints related to Stripe billing regions and current user access, suggesting updates to their billing or user management systems.

07-13URGENThacker news

A Hacker News post titled "Zig Creator Calls Spade a Spade, Anthropic Blows Smoke" is gaining significant velocity, suggesting a critical take on Anthropic by the creator of Zig.

07-13FLASHlmarena

A new model, claude-fable-5-search-v2, appeared on the LMArena leaderboard, suggesting a potential unreleased Claude variant with search capabilities.

07-13FLASHlmarena

LMArena has listed 5 new unreleased Claude and GPT models, including new 'fable-5-v2' series and a 'gpt-5.6-sol-xhigh-code' model, suggesting upcoming releases with document, text, vision, and web development capabilities. ✓ beat the announcement by 35.2h

07-19RTNweb app

ChatGPT web app updated with 9 new strings, including several related to app integrations, pricing, and privacy policies.

07-19FLASHweb app

ChatGPT web app update includes new model IDs like GPT-4.1, GPT-5, and various 'mini' and 'pro' versions, alongside internal model flags and usage cap messages.

07-19RTNweb app

Four new feature flag strings related to contentful dynamic and internal conversation rendering have been added to the ChatGPT web app frontend.

07-19RTNweb app

ChatGPT web app frontend deploy includes new strings for dynamic feature pages and internal conversation rendering.

07-19FLASHweb app

ChatGPT's web app update includes references to new model IDs like GPT-4.1, GPT-4.1-mini, GPT-4.5, GPT-5, GPT-5-mini, and GPT-5.2, alongside internal beta features and updated usage caps for GPT-4.

07-18RTNgithub

OpenAI Codex published a new pre-release for its Rust SDK, version `0.145.0-alpha.24`.

07-18FLASHweb app

OpenAI's ChatGPT web app has new strings referencing multiple unannounced GPT-4 and GPT-5 models, including 'GPT-4.1', 'GPT-4.1-mini', 'GPT-5', 'GPT-5-mini', 'GPT-5.2', 'GPT-5.3', and various 'GPT-5-pro' and 'GPT-5-reasoning' access models. It also includes new 'o3-pro', 'o4-mini', and 'o4-mini-high' model access strings, and numerous internal feature flags and endpoints.

07-18RTNweb app

ChatGPT web app frontend deploy included new strings for Cloudflare Turnstile, apps, images, and various OpenAI policy pages.

07-18RTNsitemap

OpenAI quietly published two new pages to its sitemap: one on business partners (EPAM) and another titled 'A Scorecard for the AI Age'.

07-18URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.145.0-alpha.18.

07-18RTNgithub

OpenAI Codex released a new pre-release version for its Rust integration: `rust-v0.145.0-alpha.19`.

07-18URGENTgithub

OpenAI's Codex project published a new pre-release: rust-v0.145.0-alpha.20.

07-18RTNgithub

OpenAI's Codex project published a new pre-release version: rust-v0.145.0-alpha.22.

07-18RTNgithub

OpenAI's Codex project published a new pre-release, rust-v0.145.0-alpha.23.

07-18FLASHgithub_releases+hn

OpenAI's GPT-5.6 models (Sol, Terra, Luna) received refreshed bundled instructions and an increased context window to 272,000 tokens, coinciding with an HN story about GPT-5.6 closing a 30-year gap in convex optimization.

07-18RTNgithub

OpenAI Python library v2.46.0 adds new API endpoints for managing service account API keys within organization projects.

07-18FLASHpypi

OpenAI's latest SDK introduces a new model, `gpt-4o-2024-08-06`, along with beta features for 'realtime' and 'responses' connections, and new token usage metrics.

07-18URGENTnpm

OpenAI's 6.48.0 SDK update hints at new file search tool configurations like chunking and overlap tokens, a 'ranker' option, and event streaming capabilities.

07-18RTNweb app

New strings in the ChatGPT web app hint at internal documentation features, keyboard shortcuts, and a desktop app for codex-app-prod.

07-16RTNsitemap

OpenAI quietly published four new pages to its sitemap, including two new business partner case studies and a new page about why teens deserve access to safe AI.

07-16RTNgithub

OpenAI's Codex project published a new pre-release version: rust-v0.145.0-alpha.16.

07-16FLASHweb app

OpenAI is testing new image generation styles like '80s Glam', 'Art Nouveau', and 'Classic Anime', alongside internal health data connectors for Apple Health and medical records.

07-16RTNgithub

openai/codex released a new Rust pre-release: rust-v0.145.0-alpha.15, indicating continued development on their Rust client or library.

07-15URGENTweb app

ChatGPT web app update includes new image generation styles like '80s Glam', 'Art Nouveau', and 'Classic Anime', suggesting expanded DALL-E 3 capabilities within the chat interface.

07-15FLASHweb app

OpenAI's ChatGPT web app has deployed new frontend code, revealing numerous potential new GPT model IDs like 'GPT-4.1', 'GPT-5', 'GPT-5-mini', and 'GPT-5 Thinking Mini'. The update also includes references to internal features and various image generation styles.

07-15RTNgithub

OpenAI's Codex project released a new pre-release version: rust-v0.145.0-alpha.14, suggesting ongoing development.

07-15FLASHweb app

ChatGPT's web app update hints at multiple new GPT-4 and GPT-5 model IDs, including 'GPT-4.1', 'GPT-4.1-mini', 'GPT-5 Thinking Mini', 'GPT-5.2', and 'GPT-5-mini', along with internal model search and beta features.

07-15URGENTweb app

New strings in the ChatGPT web app suggest upcoming image generation styles, including '80s Glam', 'Art Nouveau', 'Classic Anime', and 'Dramatic Headshot', along with a 'Tiranga Portrait'.

07-15FLASHmodel_catalog

LiteLLM's model catalog shows three new OpenAI GPT-5.6 model IDs hosted on Bedrock Mantle: luna, sol, and terra. ✓ beat the announcement by 6.2h

07-15URGENTweb app

OpenAI's ChatGPT web app shows strings related to medical records, Apple Health, auto top-up billing, and new tracking endpoints, suggesting potential health integrations and payment features.

07-15FLASHopenapi

Eight new OpenAI models, including gpt-5.5, gpt-5.6 variants, and gpt-realtime-2.1 models, appeared in the official API spec. ✓ beat the announcement by 7.5h

07-15FLASHweb app

OpenAI's ChatGPT frontend code has been updated, revealing multiple new and upcoming model IDs like 'GPT-4.1', 'GPT-4.1-mini', 'GPT-4o', 'GPT-5', 'GPT-5-mini', 'GPT-5 Thinking Mini', 'GPT-5.2', and 'GPT-5.3', along with internal feature flags and API endpoints.

07-15URGENThacker news

A Hacker News story titled "Codex Micro" is rapidly gaining traction, suggesting a potential new development or discussion around a compact version of OpenAI's Codex.

07-15RTNhacker news

OpenAI lost a trademark dispute at the EU court.

07-15RTNsitemap

OpenAI's sitemap quietly added 12 new URLs, mostly pointing to an unannounced 'supply/product' store with branded merchandise, plus one page on AI safety.

07-15FLASHweb app

OpenAI's ChatGPT web app was updated with many new strings, including numerous mentions of unannounced GPT-4.x and GPT-5.x model IDs, internal model flags, and new image generation styles. ✓ beat the announcement by 4.9h

07-15URGENTgithub

OpenAI Codex released a new pre-release version: rust-v0.145.0-alpha.13, indicating ongoing development. ✓ beat the announcement by 32.5h

07-15URGENTweb app

OpenAI is testing new health data features for ChatGPT, including Apple Health and medical record integration, alongside various payment and analytics updates.

07-15RTNdocs

OpenAI's changelog shows significant reorganization and expansion of API documentation, adding new sections for agents, tools, voice & audio, production, and various performance/safety aspects.

07-15RTNdocs

OpenAI's changelog mentions 'Windows app deployment', suggesting a new desktop application or improved support for Windows-based deployments.

07-15FLASHlmarena

A new model, gpt-5.6-sol-xhigh-text, has appeared on the LMArena leaderboard. ✓ beat the announcement by 14.0h

07-15URGENTgithub

OpenAI's Codex project published a new pre-release: rust-v0.145.0-alpha.12. ✓ beat the announcement by 38.9h

07-14URGENTweb app

New strings in ChatGPT's web app suggest upcoming image generation style options like '80s Glam', 'Art Nouveau', 'Classic Anime', and 'Tiranga Portrait'.

07-14URGENTnpm

OpenAI's latest SDK update (6.47.0) hints at new file search tool capabilities with configurable chunking, a new 'ranker' option, and improved event streaming for object lifecycle. It also clarifies JSON mode usage.

07-14URGENTgithub_org

OpenAI made a new public repository called 'redcard' which uses Codex to help users 'eject from work'.

07-14FLASHweb app

OpenAI's ChatGPT web app has new strings referencing several unreleased model IDs like GPT-4.1, GPT-5, GPT-5.2, GPT-5.3, GPT-4o-gdrive, and various 'mini' and 'pro' versions, alongside new internal feature flags and endpoints. ✓ beat the announcement by 22.1h

07-14RTNgithub

OpenAI's Codex project published a new pre-release: rust-v0.145.0-alpha.11, indicating ongoing development, likely for a Rust-based SDK or integration.

07-14RTNhacker news

OpenAI is mandating hardware-backed passkeys for Trusted Access Cyber members to log into ChatGPT accounts, according to a Yubico blog post.

07-14RTNhacker news

An HN post claims a bot named 'Codex' scraped the ICM website and leaked the 2026 Fields Medal winner list.

07-14URGENThacker news

Codex appears to be implementing prompt encryption and using ciphertext directly for inference.

07-14URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.145.0-alpha.10, indicating continued development.

07-14FLASHweb app

OpenAI's ChatGPT web app frontend update reveals numerous new model IDs including 'GPT-4.1', 'GPT-4o', 'GPT-5', 'GPT-5-mini', and 'GPT-5.2', alongside internal feature flags and endpoints. ✓ beat the announcement by 32.5h

07-14URGENTweb app

New strings in ChatGPT's web app suggest upcoming image generation style options like '80s Glam', 'Art Nouveau', 'Classic Anime', and 'Dramatic Headshot'.

07-14RTNgithub

OpenAI Codex published a pre-release for its Rust library, version 0.145.0-alpha.9.

07-14RTNgithub

OpenAI's Codex project published a new pre-release, rust-v0.145.0-alpha.8.

07-14URGENTweb app

ChatGPT web app update includes new image generation styles like '80s Glam', 'Art Nouveau', 'Classic Anime', 'Dramatic Headshot', and 'Tiranga Portrait'.

07-13RTNsitemap

OpenAI quietly added 35 new pages to its sitemap, primarily listing business partners in a new "partners" directory.

07-13FLASHweb app

OpenAI's ChatGPT web app updated with numerous internal model IDs like 'GPT-4.1', 'GPT-5', 'GPT-5-mini', and 'GPT-4o', alongside new feature flags and internal endpoints. This also includes hints at potential future image generation styles. ✓ beat the announcement by 43.2h

07-13FLASHweb app

ChatGPT's web app bundle includes references to new model IDs like GPT-4.1, GPT-4.1-mini, GPT-4o, GPT-5, GPT-5-mini, GPT-5.2, and GPT-5.3, alongside various 'internal' feature flags and endpoints. ✓ beat the announcement by 45.7h

07-13RTNgithub

A new pre-release, rust-v0.145.0-alpha.7, has been published for openai/codex.

07-13RTNgithub

OpenAI's Codex Rust library `rust-v0.144.2` release reverts a prompting regression related to Guardian auto-review policy and tool behavior.

07-13FLASHweb app

OpenAI's internal ChatGPT web app deploy includes new strings referencing several unreleased GPT models like 'GPT-4.1', 'GPT-4o', 'GPT-5', 'GPT-5.2', 'GPT-5.3', and various 'mini' and 'pro' versions.

07-13RTNweb app

ChatGPT web app update includes new feature flag strings, potentially related to dynamic feature pages and internal conversation rendering.

07-18FLASHgithub

The Google Gemini CLI received multiple updates, including new 'caretaker-triage' LLM orchestration features, fixes for ReAct and prompt injection loops, and various core improvements.

07-16RTNgithub

Gemini CLI published a new pre-release (v0.52.0-nightly.20260716.g3ff5ba20f) which includes a fix for handling cancelled tool responses and coalescing consecutive roles to prevent 400 Bad Request errors.

07-15FLASHmodel_catalog

Two new model IDs, "gemini-omni-flash-preview" and "gemini/gemini-omni-flash-preview," appeared in the LiteLLM model catalog, suggesting a potential new Gemini model release.

07-14RTNgithub

Google Gemini CLI published a new pre-release v0.52.0-nightly.20260714.gfa975395b, including fixes for project quota limit errors, task cancellation in a2a-server, and plan mode write policy, plus a bump to node google-auth-library version 10.9.0.

07-13URGENThacker news

xAI's Grok Build CLI appears to upload entire git repositories to a Google Cloud bucket, based on a fast-climbing HN story.

07-13URGENThacker news

A Hacker News story is rapidly gaining traction about the Grok CLI potentially uploading an entire home directory to Google Cloud Storage.

07-13RTNgithub

Google Gemini CLI published a new pre-release v0.52.0-nightly.20260713.gf354eebaf, including a fix for Code Assist tier messaging.

07-15URGENThacker news

A Hacker News story titled "Grok Build" is climbing fast, suggesting a potential open-source release or builder tool related to Grok.

07-15FLASHgithub_org

xAI has open-sourced 'grok-build', a coding agent harness with a TUI.

07-13FLASHhacker news

An HN post titled "Grok uploaded my user directory to xAI's servers" is gaining significant traction, suggesting a potential data privacy concern with Grok.

07-16RTNct_logs

A new subdomain, turn.mistral.ai, appeared in certificate transparency logs for Mistral AI, suggesting potential new infrastructure.

07-18RTNdocs

Cursor's changelog was updated with new features for their Slack integration, including multi-repo support, cross-channel workflows, and improved interaction design.

07-14URGENThacker news

An HN post titled 'Cursor 0day: When Full Disclosure Becomes the Only Protection Left' is rapidly gaining traction.

07-13URGENThacker news

Cloudflare's new "Precursor" is rapidly gaining traction on Hacker News, suggesting a significant announcement related to their developer platform.

x.ai has a new subdomain 'external-rl-env-portal.x.ai' in certificate transparency logs, suggesting new external reinforcement learning environment portal infrastructure.

Two new models, inkling-text and inkling-webdev, have appeared on the LMArena leaderboard, likely indicating unannounced stealth models being tested.

OpenRouter's API now includes a new model ID: moonshotai/kimi-k3. This often indicates an upcoming release or announcement.

An HN post titled "The LLM Critics Are Right. I Use LLMs Anyway" is gaining rapid velocity, indicating discussion around the practical use of LLMs despite their flaws.

A high-velocity Hacker News story indicates Murati's Thinking Machines released a new open-weights 975B parameter LLM named Inkling.

Microsoft pushed two new BitNet embedding models (0.6B and 270M parameters) to Hugging Face.

Nvidia pushed a new Hugging Face repository for "Nemotron-3-Embed-1B-NVFP4", suggesting a new embedding model.

Bonsai 27B, a 1-bit LLM, is gaining traction for allegedly being the first 27B-class model to run on a phone.

Nvidia released two new Nemotron-3 embedding models on HuggingFace: Nemotron-3-Embed-1B-BF16 and Nemotron-3-Embed-8B-BF16.

A Wired article detailing a leak of SFPD drone footage exposing urban surveillance is gaining rapid traction on Hacker News.