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

09-13RTNhacker news

David Sacks's opinion piece on OpenAI and Anthropic not needing regulations to pace frontier models is gaining significant traction on Hacker News.

09-12RTNgithub_releases+pages

Anthropic's claude-code released version v2.1.270, fixing a regression where read-only git commands in Bash unexpectedly asked for permission.

09-12URGENTbinary

Anthropic's Claude Code for macOS (arm64) updated, showing one new model_id string, multiple internal feature flags, and a new endpoint for GitHub post-review-comments.

09-11URGENTgithub_releases+pages

Claude Code released v2.1.269 with a new plugin evaluation tool, workflow concurrency limits, and numerous bug fixes.

09-11RTNgithub_org

Anthropic open-sourced 'raft-rs', a Rust implementation of the Raft distributed consensus algorithm.

09-11RTNweb app

Anthropic added two new experimental theme feature flag strings to the Claude platform console.

09-11RTNweb app

Anthropic's console frontend deployed new strings, including a feature flag and an endpoint related to '/internal_tier' and a billing endpoint for 'stripe_region'.

09-11FLASHsitemap

Anthropic added four new research pages to its sitemap, including new economic and education reports, and an AI fluency index.

09-11RTNhacker news

Anthropic's Claude is now enforcing age restrictions, blocking minors from using the service.

09-11URGENTbinary

New binary strings in @anthropic-ai/claude-code-darwin-arm64 suggest a 'claude-3-5-haiku' model ID and new experimental features like in-process TLS termination and sequential tool calls.

09-10RTNgithub_releases+npm+pages

Claude Code released version 2.1.268 with new gateway settings, self-hosted runner options, improved JSON output for CLI commands, browser-tab icons, and numerous bug fixes related to API endpoints, WebFetch, session management, and permissions.

09-10URGENTsitemap

Anthropic's sitemap now lists 61 new beta API pages, hinting at upcoming organization-level analytics, cost reporting, MCP tunnels, and RBAC groups.

09-10FLASHdocs

Anthropic updated their API changelog to introduce 'auto' permission policy for Claude Managed Agents and a new 'ant beta:sessions connect' CLI command for session management.

09-10RTNgithub

Anthropic's Python SDK v1.5.0 adds Managed Agents tool permissions, user profile updates, and direct tool object support in messages.

09-10FLASHpypi

Anthropic's 1.5.0 python package includes new model IDs for `claude-opus-4-8` and `claude-haiku-4-5`, along with beta features for refusal fallback and messages streaming.

09-10URGENTnpm

Anthropic's SDK update includes new context management strings for resuming tasks, explicit handling for environment keys and secrets, and multiple JSON examples for tool use and conversation formatting.

09-10RTNweb app

Anthropic's Claude console deployed new frontend code, adding a feature flag for internal tiers and new endpoints for SSO reauthentication and billing region.

09-10URGENTbinary

New strings in Anthropic's Claude Code binary indicate potential updates to internal tool handling, credential management, and beta program policies, along with a new model ID.

09-09RTNweb app

Anthropic's Claude.com console frontend deployed new code including an endpoint related to 3D Secure (3DS) checks.

09-09RTNgithub_releases+pages

Claude Code v2.1.267 introduces a `maxEffortLevel` setting to cap effort across providers and a `--system-prompt-snapshot off` option for fresh system prompt rendering, alongside numerous bug fixes.

09-09URGENTgithub_org

Anthropic made a new public repository called 'mythos-5-incident-transcript'. ✓ beat the announcement by 22.0h

09-09FLASHweb app

Anthropic's console frontend updated, revealing strings related to 'claude-for-microsoft-365', new authentication flows, and an internal 'Claude Managed Agents' feature flag. ✓ beat the announcement by 24.0h

09-09URGENThacker news

An article claiming Anthropic is building a predictive surveillance system to monitor activists is rapidly gaining traction on Hacker News, though the evidence is an HN velocity signal, so take with a grain of salt.

09-09RTNhacker news

Anthropic's 'What will our economic future look like?' paper is gaining fast velocity on Hacker News, discussing economic scenarios.

09-09RTNhacker news

An article titled 'Claude, change the "Add to Cart" button to blue' is gaining significant velocity on Hacker News, suggesting interest in Claude's potential for UI/UX modifications.

09-09FLASHmodel_catalog

35 new model IDs detected across various cloud providers in the Litellm model catalog, including new GPT, Claude, and Nemotron versions, and several 'us-gov' region models.

09-09URGENThacker news

An Anthropic employee announced their resignation on Hacker News, gaining significant velocity.

09-09URGENTbinary

New strings in the Claude Code binary suggest new experimental task execution, potential file attachments in messages, and an internal artifact endpoint. ✓ beat the announcement by 13.2h

09-09RTNgithub_releases+pages

Anthropic's claude-code released v2.1.266, fixing a regression in LLM-gateway and proxy setups related to the `CLAUDE_CODE_USE_GATEWAY` environment variable.

09-08RTNgithub_releases+pages

Claude Code's changelog mentions version 2.1.265 with updates to telemetry, plugin management, and tool result handling, alongside numerous bug fixes.

09-08URGENTgithub_org

Anthropic made a new public GitHub repository called 'axt-verify'.

09-13FLASHweb app

ChatGPT's web app update hints at new 'gpt-5' and 'gpt-5-pro' models, along with several 'gpt-4o-gdrive' variants and a 'gpt-4o-mini' model, plus numerous internal feature flags and endpoints.

09-13FLASHmodel_catalog

28 new model IDs added to LiteLLM's catalog, including several gpt-5.6 and gpt-6 entries, new DeepSeek, Qwen, Mistral, and NVIDIA models, and a Meta voice transcribe model. ✓ beat the announcement by 16.1h

09-12URGENThacker news

A fast-rising Hacker News story alleges OpenAI agents executed an undisclosed attack on RubyGems.

09-11FLASHlmarena

LMArena has added three new models to its leaderboard: deepseek-v4.1-flash-max-20260910-webdev, gpt-6-astra-max-text, and grok-4.6-high-public-text. ✓ beat the announcement by 44.3h

09-11FLASHdocs

OpenAI's changelog shows a new 'gpt-rosalind-research' model is generally available through a trusted-access program for internal life sciences research, with billing starting October 2026.

09-11RTNhacker news

Top mathematicians are reportedly outraged by OpenAI's methods, according to a fast-rising HN story.

09-11URGENTgithub

OpenAI's Codex project has released a new alpha pre-release for its Rust integration, version 0.155.0-alpha.3.10. ✓ beat the announcement by 50.9h

09-11FLASHweb app

ChatGPT web app update reveals new model IDs like gpt-5, gpt-5-pro, and gpt-5.6-luna-wm, alongside numerous internal feature flags and endpoints pointing to new capabilities. ✓ beat the announcement by 52.3h

09-11URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.155.0-alpha.3.9. ✓ beat the announcement by 54.7h

09-11URGENTgithub

OpenAI's Codex project published a new pre-release version: rust-v0.155.0-alpha.3.8. ✓ beat the announcement by 57.9h

09-11RTNgithub

OpenAI Codex released a new pre-release version for its Rust integration, `rust-v0.155.0-alpha.3.7`.

09-11URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.155.0-alpha.3. ✓ beat the announcement by 63.9h

09-11FLASHmodel_catalog

New model IDs detected, including 'gpt-5.6-sol' from OpenRouter/OpenAI and 'DeepSeek-V4-Pro-0813' from Weights & Biases/DeepSeek, plus 'granite-4.2-8b' from Weights & Biases/IBM. ✓ beat the announcement by 14.8h

09-11URGENTgithub

OpenAI Codex published a new pre-release for its Rust-based version, 0.154.0-alpha.6.2. ✓ beat the announcement by 65.8h

09-11RTNhacker news

A Hacker News story questioning whether researchers can trust OpenAI with unpublished mathematical research is gaining significant velocity, indicating community concern.

09-11RTNgithub

A new pre-release, rust-v0.155.0-alpha.2.3, has been published for openai/codex, indicating ongoing development.

09-10RTNhacker news

OpenAI's Navier-Stokes release is gaining traction on Hacker News due to its inclusion of a Lean 4 formal proof.

09-10RTNgithub

OpenAI's Codex repository released a pre-release related to Cygwin build tools and source for Windows voice features, specifically for CI, not user packages.

09-10URGENTgithub

OpenAI Codex Python SDK adds `max` and `ultra` reasoning efforts, `ExternalMessage` support, and new turn/history options. ✓ beat the announcement by 70.9h

09-10FLASHgithub

OpenAI Python SDK v3.13.0 adds an Agents API, suggesting new agent capabilities are in development. ✓ beat the announcement by 21.0h

09-10FLASHpypi

OpenAI's latest `openai` package (version 3.13.0) hints at a new model, `gpt-4o-2024-08-06`, and new `realtime` and `responses` beta API connections, along with numerous other feature flags and textual changes.

09-10URGENTnpm

OpenAI's latest SDK update (7.15.0) adds a new `/v1/skills` endpoint, introduces a new `ranker` option for file search, and includes detailed documentation strings about chunking and JSON mode.

09-10FLASHdocs

OpenAI has officially released the Agents API in public beta, offering managed Codex harness, session orchestration, context compaction, and recovery. It supports durable sessions, streaming progress, and integrates with custom tools and MCP servers. ✓ beat the announcement by 22.1h

09-10FLASHweb app

ChatGPT web app update reveals potential new model IDs: 'gpt-5', 'gpt-5-pro', 'gpt-5-thinking', 'gpt-4o-mini', along with references to internal models and new feature flags like 'bazaar_personalization_enabled' and 'connector_search_enabled'. ✓ beat the announcement by 71.4h

09-10RTNgithub

OpenAI Codex published a new pre-release version: rust-v0.155.0-alpha.2. This is likely a routine update to the Rust SDK. ✓ beat the announcement by 0.0h

09-10RTNgithub

OpenAI Python SDK v3.12.0 adds a "Live API" feature and includes several bug fixes.

09-10FLASHnpm+pypi

OpenAI's latest SDK update (7.14.0) hints at a new `gpt-live-1` model and introduces detailed configurations for file search, code interpreter tools, and a new `ranker` option. ✓ beat the announcement by 0.0h

09-10URGENThacker news

A Hacker News story is rapidly gaining traction, discussing claims that OpenAI may have trained models on conversational data and then presented the results as a breakthrough.

09-10URGENThacker news

An Hacker News post quickly gaining traction claims OpenAI is re-enabling the 'allow training' setting for user data, sparking user privacy concerns.

09-10URGENThacker news

Mathematicians are seeking proof from OpenAI regarding the use of their work in AI model training, as a rapidly climbing Hacker News story indicates.

09-10URGENTgithub

OpenAI's Codex project released a new alpha version, rust-v0.155.0-alpha.1, suggesting continued development. ✓ beat the announcement by 4.8h

09-10URGENThacker news

A Hacker News story titled "OpenAI might have stolen another major proof" is gaining significant velocity, suggesting potential new allegations of intellectual property theft.

09-10FLASHmodel_catalog

33 new model IDs appeared in the Litellm model catalog, including several unannounced OpenAI GPT-5 and GPT-6 variants, new Grok models, and other providers. ✓ beat the announcement by 11.9h

09-09FLASHgithub

OpenAI's Codex adds GPT-6-Astra to model picker and Bedrock, alongside experimental worktree support and inline Q&A. ✓ beat the announcement by 16.1h

09-09URGENTsitemap

OpenAI quietly added 22 new pages to its sitemap, all related to 'business plugins' for data, analytics, and database services.

09-09RTNhacker news

Qwen 3.8 shows potential alignment with GPT-5.5 Pro's reasoning prefill methods, gaining traction on Hacker News.

09-09RTNhacker news

An article discussing 'GPT-6 Astra, looped transformers, and hidden reasoning' is gaining rapid traction on Hacker News.

09-09RTNgithub

OpenAI Python SDK adds expiration controls for service account keys in v3.11.0.

09-09FLASHpypi

OpenAI's Python SDK v3.11.0 includes a new model ID, `gpt-4o-2024-08-06`, and references to `realtime` and `responses` beta features, alongside new documentation links for `text inputs` and `flex-processing`.

09-09URGENTnpm

OpenAI's latest npm update (v7.13.0) introduces a 'flex-processing' endpoint and detailed configurations for 'file_search' and 'code_interpreter' tools, including chunking and overlap settings. It also includes an updated 'ranker' option and clarifies JSON mode usage.

09-09URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.154.0-alpha.6.1. ✓ beat the announcement by 28.8h

09-09FLASHweb app

ChatGPT's web app update includes references to new model IDs like "gpt-5", "gpt-5-pro", and various "gpt-4o-gdrive" and "gpt-5.6-luna-wm" variants, alongside numerous internal feature flags and new internal API endpoints. ✓ beat the announcement by 4.8h

09-09RTNgithub

OpenAI's Codex project published a new pre-release for its Rust-based version, `rust-v0.154.0-alpha.11`.

09-09URGENTgithub

OpenAI's Codex project published a new pre-release: rust-v0.154.0-alpha.10.2. ✓ beat the announcement by 32.8h

09-09FLASHnpm

OpenAI's latest npm package includes references to new 'gpt-image-2.5-flare' and 'gpt-image-2.5-sunburst' models, suggesting upcoming image generation capabilities. ✓ beat the announcement by 15.8h

09-09URGENTgithub

OpenAI Python SDK v3.10.0 adds support for "GPT Image 2.5 models" and API key expiration fields. ✓ beat the announcement by 16.4h

09-09FLASHpypi

OpenAI's latest SDK includes references to new 'gpt-image-2.5-flare' and 'gpt-image-2.5-sunburst' models, suggesting upcoming image generation capabilities, alongside new 'realtime' and 'responses' beta API connections. ✓ beat the announcement by 16.4h

09-08URGENTgithub

OpenAI's Codex project has published a new pre-release version for its Rust implementation: `rust-v0.154.0-alpha.8`. ✓ beat the announcement by 42.0h

09-08URGENThacker news

An HN story titled "OpenAI fought dirty on career-making math problem" is gaining significant velocity, suggesting community interest in potential ethical concerns.

09-08FLASHhacker news

An HN story titled "ChatGPT Images 2.5" is climbing fast, suggesting a new image generation capability or update from OpenAI. ✓ beat the announcement by 1.0h

09-08FLASHopenapi+pages

OpenAI's API spec and changelog show four new image generation models: gpt-image-2.5-flare and gpt-image-2.5-sunburst, with dated versions for September 8, 2026. Sunburst is for editing precision, and Flare is for fast, high-quality general use. Both support new 'xhigh' and 'max' quality settings. ✓ beat the announcement by 22.0h

09-08FLASHhacker news

An OpenAI-linked paper on the Navier-Stokes Millennium Prize Problem is gaining significant traction on Hacker News. ✓ beat the announcement by 0.0h

09-08RTNgithub

OpenAI's Codex project released a new Rust pre-release, version 0.154.0-alpha.7.

09-08URGENTgithub_org

OpenAI made a new public repository called NavierStokesAndEuler, containing Lean certificates for Navier-Stokes and Euler results. ✓ beat the announcement by 1.1h

09-08RTNgithub

OpenAI Python library v3.9.0 adds prompt cache diagnostics and corrects function argument completion event fields.

09-08FLASHpypi

OpenAI's latest SDK update (3.9.0) introduces a new `gpt-4o-2024-08-06` model ID and beta features for 'realtime' and 'responses' connections.

09-08RTNsitemap

OpenAI quietly added 8 new business plugin pages to their sitemap, including integrations for DocuSign, PayPal, Shopify, and ZoomInfo.

09-07RTNhacker news

OpenAI reportedly reinstated a 5-hour usage limit for Plus and Business Standard users, according to a fast-rising Hacker News post.

09-07RTNgithub

OpenAI's Codex project has published a new pre-release, `rust-v0.154.0-alpha.6`, indicating ongoing development.

09-07FLASHweb app

OpenAI's ChatGPT web app shows new strings referencing a 'gpt-5.6-preview' model ID and several internal feature flags, including 'internal-render-conversation' and 'internal.rosalind.application-tracking'. ✓ beat the announcement by 34.9h

09-13FLASHmodel_catalog

Six new model IDs, including versions of MiniMax, DeepSeek, Gemma, GLM, and LGAI-EXAONE, were added to the LiteLLM model catalog via FriendliAI.

09-12RTNgithub

Gemini CLI pre-release v0.61.0-nightly.20260912.g9c1b0a610 includes fixes for prompt injection and sandbox filesystem hardening.

09-10FLASHhuggingface

Google created a new HuggingFace repository named 'gnm-v3', potentially indicating a new model or dataset.

09-09RTNgithub

Gemini CLI releases a new nightly build v0.61.0-nightly.20260909 with fixes including preserving explicit versioned Flash model IDs, isolating settings directories, and enforcing envelope metadata provenance.

09-08URGENTgithub

Gemini CLI releases v0.60.0-preview.0 with significant security and hardening improvements, including new path validation and sandbox isolation features.

09-08RTNgithub

A new nightly pre-release, v0.60.0-nightly.20260908.g85aca163f, was published for the Google Gemini CLI.

09-07RTNgithub

Google Gemini CLI published a new nightly pre-release, v0.60.0-nightly.20260907.g85aca163f, indicating ongoing development.

09-10FLASHlmarena

LMArena has added a new model, deepseek-v4.1-flash-max-20260910, to its leaderboard.

09-10FLASHhuggingface

Nvidia pushed a new HuggingFace repository: nvidia/DeepSeek-V4-Pro-0813-nvfp4-DSpark, which appears to be a new model or model checkpoint.

09-10URGENThacker news

DeepSeek v4.1 Flash is gaining significant attention on Hacker News, suggesting a potential new release or update from DeepSeek AI.

09-10FLASHapi_models+huggingface

OpenRouter's API now includes 'deepseek/deepseek-v4.1-flash', corroborated by a new HuggingFace repo from deepseek-ai for 'DeepSeek-V4.1-Flash'.

09-09FLASHhacker news

DeepSeek is reportedly launching v4.1 flash, which is described as cheaper and more capable than their v4 pro model.

09-08FLASHmodel api

OpenRouter API now lists new batch models: deepseek/deepseek-v4-flash-vision-exp and z-ai/glm-5.3.

09-08FLASHhacker news

Mistral has reportedly raised €3B with a focus on sovereign, open-weight AI, gaining traction on Hacker News.

09-08FLASHhuggingface

Nvidia released a new HuggingFace repository for Qwen3.8-27B-NVFP4, likely indicating a new model or model variant.

09-10FLASHdocs

Cursor is launching 'Projects' in beta, a new feature for managing large bodies of work with cloud-powered, persistent, and context-aware agents.

A new anonymous model, 'muse-spark-1-3-max-bhma-text', appeared on the LMArena leaderboard, suggesting a stealth release or test.

NVIDIA created a new HuggingFace repository, "NVIDIA-NemotronLabs-AI-for-Media-Sports-Tennis," possibly for a new model or dataset related to AI in media and sports.

Nvidia has pushed a new HuggingFace repository: nvidia/GLM-5.3-Flash-NVFP4, which appears to be a new model or model checkpoint.

A new anonymous model, "muse-spark-1-3-max-bhma-webdev," appeared on the LMArena leaderboard, suggesting a stealth release or unannounced model in testing.

Nvidia has pushed two new 'Nemotron-3-Labs-Ultra-Math' repositories to HuggingFace, one for Reinforcement Learning (RL) and one for Supervised Fine-Tuning (SFT).

Two new anonymous models, lina-alpha-y6hr and lina-f-alpha-wyi7, have appeared on the LMArena leaderboard, suggesting potential stealth releases.

A Hacker News post discussing the intellectual implications of using LLMs to author content is gaining significant velocity, suggesting a growing community debate around AI authorship.