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

08-08URGENTbinary

@anthropic-ai/claude-code-darwin-arm64 2.1.226 native binary: 13 new strings (10 endpoint, 3 feature_flag, 3 sentence) — verify

08-08FLASHgithub_releases+pages

anthropics/claude-code published a release: v2.1.226 — corroborated across 3 sources (github_releases, pages)

08-07RTNgithub

anthropics/anthropic-sdk-python published a release: v0.121.0

08-07FLASHpypi

anthropic 0.121.0 shipped: 49 interesting new strings (49 sentence, 2 model_id, 2 feature_flag)

08-07URGENTnpm

@anthropic-ai/sdk 0.116.0 shipped: 33 interesting new strings (33 sentence, 1 feature_flag)

08-07RTNweb app

Anthropic's Claude platform console deployed new frontend code with new API endpoints related to billing, organization workspaces, and API key management.

08-07FLASHgithub_releases+pages

Claude Code v2.1.224 introduces self-hosted environments for Team and Enterprise plans, allowing users to run Claude Code sessions on their own machines or containers. It also adds new plugin installation options and enhanced cross-session messaging capabilities for macOS and Linux.

08-07FLASHbinary

Anthropic's Claude-code binary shows new strings related to managed agents, multi-agent orchestration, self-hosted runners, and webhook delivery, suggesting upcoming features for advanced AI workflows.

08-06RTNweb app

Anthropic's Claude platform console shows new API key management and workspace assignment endpoints, plus a KYC status check for organizations.

08-06FLASHmodel api

OpenRouter API added 30 new model IDs, including potential new batch versions for Claude Sonnet 4.6, Moonshot Kimi K2.7 Code, NVIDIA Nemotron-3, and several unannounced OpenAI models like GPT-5, GPT-5.2 to 5.6, and O1 to O4.

08-06RTNbinary

Anthropic's Claude Code binary updated with new strings suggesting internal tooling, potential team memory features, and feedback mechanisms.

08-06FLASHgithub_releases+npm+pages

anthropics/claude-code published a release: v2.1.223 — corroborated across 3 sources (github_releases, npm, pages)

08-06URGENTweb app

Anthropic Console frontend deploy [e3935b96]: 13 new strings (12 endpoint, 2 feature_flag) — verify

08-05FLASHdocs

Anthropic has retired the Claude Opus 4.1 model, recommending an upgrade to Claude Opus 5. Researchers can request continued access via the External Researcher Access Program.

08-05RTNsitemap

Anthropic added three new pages related to 'inference hooks' to their Claude platform documentation sitemap.

08-05FLASHdocs

Anthropic adds Inference Hooks in beta for Enterprise organizations, allowing security server integration for prompt governance across Claude products.

08-05RTNweb app

Anthropic's platform console code includes new strings pointing to 'tunnels' and 'certificates' beta endpoints, suggesting new secure connection features.

08-05URGENTbinary

A new model ID, `@anthropic-ai/claude-codez3`, was found in the latest `claude-code` binary, alongside new endpoints and a background task feature flag.

08-05RTNgithub_releases+pages

Anthropic's claude-code released v2.1.222, improving isolation, fixing tool restrictions, and removing the 'ultraplan' feature.

08-05RTNweb app

Anthropic's Claude platform console deployed a new frontend with a single new endpoint related to explaining grants within organizations and workspaces.

08-04RTNweb app

Anthropic's Claude platform console shows new strings for internal tier settings, feature flags, and KYC status, suggesting updates to org management and user verification.

08-04RTNhacker news

A new open-source project, ADLC Team Skills, is gaining traction for enabling agent skills to enforce team coding standards with Claude Code and Codex.

08-04FLASHbinary

Anthropic's Claude-Code native binary update reveals new model IDs like `@anthropic-ai/claude-code1` and `claude-3`, alongside numerous internal feature flags and ingest dashboard endpoints.

08-04URGENTgithub_releases+pages

Claude Code v2.1.221 introduces a new Focus view for VSCode, improved sandbox credential handling with a 'mask' mode for Linux/WSL, and a prompt-audit subcommand for the claude-api skill.

08-04RTNweb app

Anthropic's Claude platform console deployed new frontend code with new API endpoints related to organizations, billing, KYC, feature settings, and data sharing.

08-03FLASHdocs

Anthropic's API changelog shows an unannounced 'Dreams' research preview for Claude Opus 5 and a new Compliance API Activity Feed for AWS.

08-03URGENTweb app

Anthropic's console updated with new API endpoints, including `workspaces`, `prepaid/credits`, `kyc_status`, and `supported_regions`.

08-08RTNweb app

ChatGPT web app frontend deploy [87084475]: 4 new strings (4 feature_flag) — verify

08-08RTNweb app

ChatGPT web app frontend deploy [b952a4b6]: 4 new strings (4 feature_flag) — verify

08-08FLASHweb app

ChatGPT web app frontend deploy [87084475]: 305 new strings (159 endpoint, 125 feature_flag, 22 model_id, 16 sentence) — verify

08-08RTNweb app

ChatGPT web app frontend deploy includes new strings for images, plugins, OpenAI help, a local lightweight web, and pricing pages.

08-08FLASHweb app

ChatGPT web app frontend deploy [87084475]: 309 new strings (174 endpoint, 118 feature_flag, 18 model_id, 16 sentence) — verify

08-08URGENTgithub

openai/codex published a pre-release: rust-v0.148.0-alpha.4

08-08URGENTgithub

openai/codex published a pre-release: rust-v0.148.0-alpha.5

08-08FLASHweb app

ChatGPT web app frontend deploy [87084475]: 23 new strings (17 feature_flag, 4 sentence, 3 model_id, 3 endpoint) — verify

08-07FLASHweb app

ChatGPT web app update includes new strings referencing "GPT-5" for smarter, faster responses, along with several internal health-related feature flags and new endpoints.

08-07FLASHweb app

ChatGPT web app updated with potential internal model access and a new 'gpt-4o-mini-filled' model ID, along with internal feature flags.

08-07FLASHweb app

ChatGPT web app frontend deploy [ff1c9707]: 12 new strings (8 feature_flag, 3 endpoint, 2 sentence, 1 model_id) — verify

08-07FLASHweb app

ChatGPT web app frontend deploy [7322bc0e]: 300 new strings (169 endpoint, 114 feature_flag, 18 model_id, 15 sentence) — verify

08-07URGENTdocs

OpenAI's changelog shows updates related to X.509 certificate support, a "Remote" feature for the ChatGPT desktop app, and significant restructuring and additions within the Codex Security documentation.

08-07RTNgithub

OpenAI's Codex project released a new pre-release version, rust-v0.148.0-alpha.1, indicating ongoing development, possibly with new features or improvements.

08-07URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.148.0-alpha.2.

08-07URGENTgithub

Codex `rust-v0.147.0` introduces portable agent plugins, conversation organization, `--approve-for-me` CLI flag, and Amazon Bedrock cached search.

08-07RTNweb app

A ChatGPT web app frontend deploy shows a new GPT-4 usage cap message, hints at internal features like 'Dogfood' and 'Beta', and new image picker styles.

08-07FLASHmodel_catalog

A new model ID, replicate/openai/gpt-oss-20b, has been added to the LiteLLM model catalog.

08-06FLASHweb app

ChatGPT web app frontend deploy [d2443a97]: 260 new strings (130 endpoint, 111 feature_flag, 20 model_id, 12 sentence) — verify

08-06RTNweb app

ChatGPT web app frontend deploy [394bf7c4]: 8 new strings (8 endpoint) — verify

08-06RTNweb app

ChatGPT web app frontend deploy [fc9f5239]: 4 new strings (4 feature_flag) — verify

08-06URGENTgithub

OpenAI's Codex project has released a new pre-release version: rust-v0.147.0-alpha.13, suggesting ongoing development in the Rust codebase.

08-05URGENTweb app

ChatGPT web app shows a new string indicating a GPT-4 usage cap has been reached, suggesting a possible tiered access or throttling mechanism.

08-05RTNgithub

OpenAI's Codex project published a new pre-release for its Rust-based version, 0.147.0-alpha.12.

08-05RTNhacker news

HN velocity: "Beating GPT-5.6 Sol on retrieval with 100x cheaper open models" (44 pts/hr)

08-05URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.147.0-alpha.6.5.

08-05RTNweb app

ChatGPT's web app was updated with 51 new strings, including a GPT-4 usage cap message, several feature flags, and numerous external service endpoints.

08-05FLASHdocs

OpenAI's changelog shows new GPT-5.6 models (Sol, Terra, Luna) with long-context fast mode, and API key filtering in usage dashboards, dated August 2026.

08-05RTNgithub

OpenAI Codex published a pre-release for its Rust binding, version 0.147.0-alpha.11.

08-05FLASHweb app

ChatGPT's web app update includes references to 'GPT-5', 'gpt-4l', and several 'gpt-4o-gdrive' models, alongside numerous internal feature flags and endpoints.

08-05RTNgithub

OpenAI's Codex project released rust-v0.146.1, which includes safer automatic-review defaults for cyber-capable models and better explanations for permission changes.

08-05FLASHweb app

ChatGPT web app update includes references to new internal model IDs like 'GPT-5', 'gpt-4o-gdrive', and 'gpt-5-pro', alongside numerous internal feature flags and endpoints.

08-05URGENTgithub

OpenAI's Codex project published a new pre-release, rust-v0.147.0-alpha.10, indicating ongoing development in Rust.

08-05RTNweb app

ChatGPT web app update includes new strings referencing image features, plugins, and various OpenAI help and policy pages, including Korean language support.

08-04RTNgithub

OpenAI's Codex project published a new pre-release for its Rust-based version, 0.147.0-alpha.6.4.

08-04FLASHweb app

ChatGPT web app update reveals multiple new model IDs, including "GPT-5", "gpt-4l", and several "gpt-4o-gdrive" variants, alongside new internal feature flags and endpoints.

08-04URGENTgithub

OpenAI's Codex project published a new pre-release, `rust-v0.147.0-alpha.6.3`.

08-04RTNweb app

ChatGPT web app update includes new feature flags for dynamic and internal conversation rendering, and feature pages.

08-04FLASHweb app

OpenAI's ChatGPT web app shows new strings referencing 'GPT-5' and several 'gpt-4o-gdrive' variants, alongside numerous 'internal' model and feature flags.

08-04URGENThacker news

An HN story is rapidly gaining traction regarding Apple's claims that more ex-employees may have transferred confidential data to OpenAI.

08-04RTNgithub

OpenAI Codex published a new pre-release, rust-v0.147.0-alpha.6.1, indicating continued development in its Rust implementation.

08-04URGENTweb app

ChatGPT's web app got an update with strings hinting at new features like "early-access" and "work-internal-model-search", along with a GPT-4 usage cap message.

08-04FLASHweb app

OpenAI's ChatGPT web app has new frontend code with multiple mentions of 'GPT-5' and internal model references like 'GPT-5-pro' and 'GPT-5-reasoning-mini'.

08-04URGENThacker news

An article titled 'Apple is getting this wrong' is gaining significant traction on Hacker News, suggesting a potential critique from OpenAI regarding Apple's strategy or recent announcements.

08-04FLASHweb app

OpenAI's ChatGPT web app update reveals multiple new GPT-5 model IDs, including 'GPT-5-pro', 'GPT-5-reasoning', 'GPT-5.1', and 'GPT-5.2', alongside internal model picker features and experimental conversation endpoints.

08-03RTNgithub

OpenAI's Codex project released a new pre-release version for its Rust component: `rust-v0.147.0-alpha.6`.

08-03URGENTgithub

OpenAI's Python library release v2.53.0 adds `gpt-5.5` and tool name/namespace to Response types, indicating potential future model and tool updates.

08-03FLASHpypi

OpenAI's latest client library adds a new model ID `gpt-4o-2024-08-06` and hints at new 'realtime' and 'responses' beta API connections, along with JSON mode improvements and better fine-tuning recommendations.

08-03RTNnpm

The OpenAI NPM package v7.4.0 adds 'ranker' options like 'default_2024_08_21', introduces chunking controls for file_search, clarifies JSON mode usage, and adds event emission details.

08-03FLASHweb app

ChatGPT's web app update includes numerous references to "GPT-5" and several variations, along with new internal model names and features like "Internal models" search.

08-03FLASHweb app

New strings in ChatGPT's web app hint at a 'gpt-5-5-instant' model and experimental features like autocompletions, trending suggestions, and 'magic edit'.

08-03RTNhacker news

An OpenAI blog post titled 'Ten advances in mathematics and theoretical computer science' is gaining traction on Hacker News.

08-03URGENTgithub

OpenAI's Codex project published a new pre-release version `rust-v0.147.0-alpha.1.2`, indicating ongoing development.

08-03FLASHpypi

OpenAI's latest client update, 2.52.1, includes a new `gpt-4o-2024-08-06` model ID and introduces `realtime` and `responses` beta API connections, suggesting new streaming or interactive capabilities.

08-03URGENThacker news

OpenAI's super PAC is reportedly funding an AI-generated news site that attacks critics of the AI industry, gaining significant traction on Hacker News.

08-09URGENTgithub

google-gemini/gemini-cli published a pre-release: v0.56.0-nightly.20260809.gcf22ac7e8

08-08URGENTgithub

google-gemini/gemini-cli published a pre-release: v0.56.0-nightly.20260808.gcf22ac7e8

08-07RTNgithub

Gemini CLI has new pre-releases (v0.56.0-nightly, v0.55.0-preview) and a patch release (v0.54.4), indicating ongoing development and bug fixes.

08-06FLASHgithub

google-gemini/gemini-cli published a pre-release: v0.55.0-preview.1 — corroborated across 3 sources (github_releases)

08-04FLASHmodel_catalog

Two new Gemini robotics models, 'gemini-robotics-er-1.6-preview' and 'gemini-robotics-er-2-preview', have appeared in the LiteLLM model catalog.

08-04URGENTdocs

Cursor introduces new Google Workspace plugins for coding agents, enabling direct access to Gmail, Drive, Calendar, Docs, and Sheets for reading, writing, and acting without leaving the IDE.

08-03URGENTdocs

Cursor introduces new Google Workspace plugins for Drive, Gmail, Calendar, Docs, Sheets, and Chat.

08-03FLASHlmarena

A new model, gemini-3.5-flash-high-public, has appeared on the LMArena leaderboard, potentially indicating an upcoming release.

08-03RTNgithub

Gemini CLI published a new nightly pre-release, v0.55.0-nightly.20260803.gf47d6c6f7.

08-07URGENThacker news

HN velocity: "DeepSeek V4 Flash 0731" (98 pts/hr)

08-06RTNhacker news

HN velocity: "DeepSeek planning to significantly raise prices" (35 pts/hr)

08-04RTNhacker news

A DeepSeek V4 Flash implementation is gaining traction on Hacker News for running on a single AMD MI300X.

08-04FLASHhacker news

Mistral has released Shieldstral, a new 3B open-weights model designed for multimodal moderation, gaining significant traction on Hacker News.

08-04FLASHhuggingface

Mistral AI published a new Hugging Face repository, mistralai/Shieldstral-1.0-3B, which appears to be a new model.

08-03FLASHmodel api

OpenRouter API shows new model ID: qwen/qwen3.8-max, indicating a likely upcoming release.

HN velocity: "How I use LLMs to learn complex topics" (64 pts/hr)

LMArena: 2 new models on the leaderboard (horseshoe-zflu, porcelain-k9cy) — anonymous, likely a stealth model

HN velocity: "Responding to the next frontier of critical cyber capabilities" (44 pts/hr)

Three new models (muse-spark-1.2-xhigh-text, muse-spark-1.2-xhigh-vision, muse-spark-1.2-xhigh-webdev) appeared on the LMArena leaderboard, likely representing a stealth release.

A new model, kimi-k3-v3-text, has appeared on the LMArena leaderboard.

x.ai: 2 new subdomains appeared in certificate-transparency logs

HN velocity: "Born Against, or why hobby programming communities are against LLM usage" (41 pts/hr)

A paper titled "Position: LLMs Can't Jump" is gaining traction on Hacker News, suggesting limitations in LLM capabilities.

Two new anonymous models, "cosmos3-super-agentic-nvcf-v2" and "cosmos3-super-nvcf-v2," appeared on the LMArena leaderboard, likely indicating unannounced models currently in testing.

Nvidia created a new HuggingFace repository named 'corrdiff-cosmo-era5', potentially indicating a new model or dataset related to correlation diffusion.

Two new anonymous models, "thunbergia-alpha-jz4q" and "zen-bear-v14r-z42u," have appeared on the LMArena leaderboard, suggesting potential stealth testing of unreleased models.

Nvidia has pushed two new HuggingFace repositories: NVIDIA-Nemotron-Parse-2.0 and NVIDIA-NemotronLabs-VoiceChat-11B.

Two new anonymous models, "hailuo-3-i2v" and "hailuo-3-text-to-video," have appeared on the LMArena leaderboard, indicating potential stealth development of image-to-video and text-to-video capabilities.

A Hacker News story titled "LLMs reward expertise" is gaining rapid traction, suggesting a growing community interest in how specialized knowledge impacts large language model performance and utility.

A trending Hacker News article suggests manually retyping LLM-generated code to prevent cognitive debt.

A critical CVE was issued for a hallucinated SQLite vulnerability, which gained significant attention on Hacker News.

Three new anonymously named models (kinsley-5jmg, kinsley-mrp8, thunbergia-alpha-9e3w) have appeared on the LMArena leaderboard, potentially indicating unreleased stealth models being tested.