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

09-06RTNbinary

New strings in the @anthropic-ai/claude-code-darwin-arm64 binary suggest internal task management, handling rejected beta features, and clearer instructions on how the AI should interpret system notifications versus user messages.

09-05URGENTbinary

Anthropic's Claude-code binary includes new strings hinting at potential 'dangerous tool use' safeguards, internal event tracking, and marketplace integration.

09-04RTNgithub

Anthropic's Python SDK v1.4.0 adds Claude Tag categories to usage reports, new types for organization compliance, and workspace ID support.

09-04FLASHpypi

Anthropic's 1.4.0 Python package includes references to new models, "claude-opus-4-8" and "claude-haiku-4-5", and introduces a beta refusal fallback middleware feature.

09-04URGENTnpm

Anthropic's SDK update includes new example JSON for tool use, structured content, and continuation summaries, plus a `myContext` request option.

09-04RTNweb app

Anthropic's Claude platform deployed new frontend code, including a new SVG endpoint that may relate to UI updates or asset loading.

09-04URGENTgithub_releases+pages

Claude Code v2.1.261 adds organization policy status, increased output limits, file-based subagent prompts, and a skill-doctor for pruning unused skills.

09-04URGENThacker news

An Anthropic research paper on formalizing Fermat's Last Theorem is gaining rapid traction on Hacker News, indicating significant interest in their mathematical formalization work.

09-04RTNhacker news

An HN story is rapidly gaining traction discussing the recent unacknowledged outages at OpenAI and Anthropic, with no clear explanation provided by either company.

09-04URGENTgithub_org

Anthropic made a new public GitHub repository named 'fermats-last-theorem'.

09-04RTNweb app

Anthropic's Claude platform added new API endpoints related to billing, workspaces, and a contact sales link.

09-04URGENTbinary

Anthropic's `claude-code-darwin-arm64` binary update includes references to new `claude-sonnet-4-6` model pricing, experimental task execution, and internal debugging/feature flags, suggesting upcoming model iterations and development features.

09-04RTNgithub_releases+pages

Claude Code (v2.1.260) adds a diff panel, prompt-cache miss causes to `/cost`, and `/reload-plugins` for headless sessions, along with numerous fixes.

09-03RTNweb app

Anthropic's Claude platform console deployed new frontend code, including a new billing endpoint, a Slack archive URL, and an SVG image data URL.

09-03FLASHgithub_releases+model_catalog+openapi

OpenAI's API spec and internal catalog show a new model, "gpt-6-astra," along with numerous other unreleased models from various providers like Claude, Gemini, and DeepSeek. ✓ beat the announcement by 0.0h

09-03RTNweb app

Anthropic's Claude platform console deployed new frontend code including a 'docs-show-beta' feature flag and new endpoint strings.

09-03URGENThacker news

High HN velocity on a post asking if simultaneous outages for OpenAI, Claude, and Grok are a coincidence.

09-03URGENThacker news

An HN story is rapidly gaining traction, reporting that ChatGPT, Claude, and Grok are currently experiencing downtime.

09-03URGENThacker news

High HN velocity on a Claude status page incident about elevated errors for multiple models.

09-03FLASHweb app

Anthropic's console frontend deployed new code, revealing new Claude model IDs and API endpoints.

09-03URGENTbinary

Anthropic's Claude-code binary update shows new strings related to internal tool call serving, session management, and potential 'pins_enabled' feature flags, alongside an interesting model_id path.

09-03FLASHlmarena

A new model, claude-fable-5.1-max-text, has appeared on the LMArena leaderboard, potentially indicating an upcoming release from Anthropic.

09-02URGENTgithub_releases+npm+pages

Claude Code v2.1.259 released with new `managedMcpServers` setting, `--permission-prompts none` for headless hosts, GitLab MR recognition, and various bug fixes.

09-02RTNdocs

Anthropic updated Claude Enterprise API docs, clarifying that `anthropic-version` header is required for Admin, Analytics, and Compliance APIs, similar to other Claude API endpoints.

09-02FLASHweb app

Anthropic's Claude platform deployed new frontend strings referencing two new model IDs: 'claude-fable-5-1' and 'claude-mythos-5-1', alongside new feature flags and API endpoints related to compliance, workspaces, and feedback. ✓ beat the announcement by 35.1h

09-02FLASHlmarena

A new model, claude-fable-5.1-max-webdev, appeared on the LMArena leaderboard, suggesting a potential unreleased Claude variant focused on web development.

09-02RTNhacker news

Aisle discovered six curl CVEs, while OpenAI and Anthropic's security reviews apparently found none.

09-02URGENTgithub_org

Anthropic released a new public GitHub repository, 'commerce-agents', featuring reference blueprints and examples for building shopping and merchant agents using Claude.

09-02FLASHmodel api

OpenRouter API shows new model id anthropic/claude-fable-5.1:batch, suggesting a new Anthropic batch processing model.

09-02RTNbinary

A new endpoint `bun.report` was detected in the `anthropic-ai/claude-code-darwin-arm64` binary. This could indicate new telemetry or integration with Bun, a JavaScript runtime.

09-01RTNgithub_releases+pages

Claude Code v2.1.258 was released, fixing launch issues on macOS 12 and a bug with remote/scheduled sessions failing due to empty content messages.

09-01FLASHsitemap

Anthropic's sitemap quietly published 421 new pages, including references to unannounced models like Fable 5, Mythos 5, Opus 4.5 through 4.8, Opus 5, Haiku 4.5, Sonnet 4.5, and Sonnet 5, along with their migration guides and release notes, predominantly in German.

09-01RTNweb app

Anthropic's Claude platform deployed new frontend code, including a new billing API endpoint related to Stripe regions.

09-01FLASHmodel_catalog

LiteLLM's model catalog has added 111 new model IDs across multiple cloud providers, including new versions of Amazon's Nova, Anthropic's Claude Fable, DeepSeek, GLM, Gemma, and numerous Qwen models, as well as ElevenLabs Scribe v2 and GigaChat-2.

09-01FLASHbinary

New model IDs 'claude-fable-5-1' and 'claude-mythos-5-1' detected in Anthropic's Claude Code binary, alongside numerous internal and experimental feature flags.

09-01FLASHhn+sitemap

Anthropic quietly published two new pages to its sitemap, including "Claude Fable and Mythos 5.1" and "Enterprise Frontier Safeguards," with the former quickly gaining traction on Hacker News. ✓ beat the announcement by 0.0h

09-01RTNgithub

Anthropic's Python SDK v1.3.0 introduces beta user profiles with `external_user_onboarded_at`, organization compliance settings, user-profile ordering, and memory-store/toolset schema updates.

09-01FLASHnpm+pages+pypi+sitemap

Anthropic's SDK update 0.123.0 reveals new model IDs "claude-fable-5-1" and "claude-mythos-5-1", alongside a "thinking-binding-controls-2026-08-01" beta feature flag and an organization compliance settings endpoint. ✓ beat the announcement by 0.0h

09-01FLASHgithub_releases+npm+pages

Claude Code v2.1.257 introduces Claude Fable 5.1 as the new default Fable model with 1M context, new time and timezone settings, and various security and workflow improvements. ✓ beat the announcement by 0.0h

09-01FLASHmodel api

OpenRouter API shows new Anthropic model: anthropic/claude-fable-5.1, suggesting a potential unannounced Claude variant. ✓ beat the announcement by 0.0h

09-01RTNbinary

A new internal feature flag and error message related to a 'bridge:repl' component appeared in the Claude code for Darwin ARM64, suggesting internal development around a new REPL or interactive environment.

08-31URGENTdocs

Anthropic's Claude API changelog shows numerous recent updates, including new SDKs, agent features like budget setting and memory, and model deprecations, all dated in the future.

08-31RTNweb app

Anthropic's Claude platform deployed new frontend code, adding new internal tier and notification center endpoints, and new billing, challenge redirect, and return URL strings.

08-31RTNgithub_releases+pages

Claude Code released v2.1.252, fixing issues with Bash commands on Macs, 'always allow' settings, stalled Remote Control sessions, and background task notifications exceeding API limits.

09-06URGENThacker news

OpenAI's "An Alien Mind" is gaining rapid traction on Hacker News, suggesting a new blog post or announcement.

09-06RTNweb app

ChatGPT web app updated with 19 new endpoint strings, including references to Korean policies and a new SDK.

09-06RTNweb app

A new feature flag `generate_trending_suggestions` was detected in the ChatGPT web app bundle, suggesting potential development of a trending conversation suggestion feature.

09-06FLASHmodel_catalog

LiteLLM's model catalog shows 257 new model IDs, including new Azure GPT models (e.g., gpt-6-astra, gpt-realtime-2.1), Databricks Gemini/GPT models, and several new Qwen, Mistral, and Cohere models, hinting at upcoming releases and model updates across various providers. ✓ beat the announcement by 14.3h

09-05FLASHlmarena

A new model, gpt-6-astra-max-code-codex-harness, appeared on the LMArena leaderboard, potentially indicating an upcoming release. ✓ beat the announcement by 22.0h

09-05RTNgithub

OpenAI's Codex project received a hotfix release, version rust-v0.153.4, addressing visibility and guidance issues for a bundled model named Astra.

09-04FLASHmodel api

OpenRouter's API shows new model IDs `openai/gpt-6-astra-pro:batch` and `openai/gpt-6-astra:batch`, suggesting upcoming batch inference capabilities for new GPT-6 Astra models. ✓ beat the announcement by 4.2h

09-04FLASHapi_models+github_releases

New model IDs `openai/gpt-6-astra` and `openai/gpt-6-astra-pro` appeared in OpenRouter's API, and GPT-6-Astra was added to Amazon Bedrock in an OpenAI Codex release. ✓ beat the announcement by 7.2h

09-04URGENThacker news

A new OpenAI agent message board called 'Collusion Wiki' is rapidly gaining traction on Hacker News, suggesting a potential unannounced agent-related development from OpenAI.

09-04URGENTgithub

OpenAI's Codex project published a new pre-release, `rust-v0.154.0-alpha.3`.

09-04RTNgithub

OpenAI's Codex project updated the description for the GPT-6-Astra Fast tier to accurately reflect a "2x speed, increased usage" benefit, correcting a previous typo of "1.5x speed."

09-03URGENTgithub

OpenAI's Codex project published a new pre-release for its Rust-based version, `rust-v0.154.0-alpha.2`. ✓ beat the announcement by 1.0h

09-03URGENThacker news

A Hacker News post titled "OpenAI's GPT-6 Astra on ARC-AGI-3" is gaining significant velocity, hinting at a potential new OpenAI model or benchmark. ✓ beat the announcement by 0.0h

09-03RTNhacker news

An OpenAI-linked project titled "Prime Gaps at Most 186" is gaining traction on Hacker News, hinting at potential work in number theory or mathematical research.

09-03FLASHgithub_releases+npm+pages+pypi+sitemap

OpenAI's latest SDK update (7.10.0) introduces references to unannounced models like 'gpt-6-astra', 'gpt-5.6', and 'gpt-6', alongside a new 'gpt-4o-2024-08-06' model, and features like a safety alerts endpoint and advanced file search chunking options. ✓ beat the announcement by 1.0h

09-03FLASHhacker news

High HN velocity suggests OpenAI is rolling out a new model, "GPT-6 Astra," corroborated by multiple HN posts. ✓ beat the announcement by 2.0h

09-03FLASHhacker news

An Axios article titled "Welcome to the AGI era," OpenAI says as GPT-6 Astra debuts, is gaining rapid velocity on Hacker News, suggesting high interest in a potential unannounced OpenAI model. ✓ beat the announcement by 2.9h

09-03URGENTgithub_org

OpenAI made two new public repositories related to formalizing bounds and certificates for prime gaps using Lean.

09-03URGENThacker news

OpenAI's Codex service appears to be experiencing an outage, indicated by a rapidly climbing Hacker News story. ✓ beat the announcement by 7.9h

09-03URGENTgithub

OpenAI's Codex project has released a new alpha version, rust-v0.154.0-alpha.1, indicating ongoing development in its Rust implementation. ✓ beat the announcement by 12.9h

09-03RTNweb app

ChatGPT web app is testing a new feature flag for 'generate_trending_suggestions' in conversations.

09-03RTNweb app

A new feature flag `generate_trending_suggestions` was added to ChatGPT's web app, suggesting potential upcoming functionality for trending conversation topics.

09-03URGENTgithub

OpenAI's Codex Rust library received an update (v0.153.0) with new features including Vim mode undo/redo, plugin CLI improvements, and an experimental context management feature for ChatGPT Plus/Pro/Pro Lite sessions using the Codex backend. ✓ beat the announcement by 20.1h

09-03URGENThacker news

A METR report on a hacking incident involving OpenAI and Hugging Face is gaining significant traction on Hacker News.

09-03RTNgithub

OpenAI's Codex project published a new pre-release for its Rust integration, version `rust-v0.153.0-alpha.5.1`.

09-02RTNdocs

OpenAI updated its API error codes to distinguish between rapid traffic increases and temporary model overload, introducing 'slow_down' and 'server_is_overloaded' codes.

09-02URGENTgithub

OpenAI Codex published a new pre-release version rust-v0.153.0-alpha.6. ✓ beat the announcement by 34.8h

09-02RTNweb app

ChatGPT web app deployed new frontend code, including a feature flag for 'generate_trending_suggestions'.

09-02RTNgithub

A new pre-release, rust-v0.153.0-alpha.5, has been published for openai/codex.

09-02FLASHnpm+pypi

OpenAI's latest SDK update (v7.9.0) adds new model IDs 'gpt-4o-2024-08-06' and 'gpt-5.6-sol', alongside new features for file search tool controls, JSON mode, and event streaming.

09-02RTNgithub

OpenAI Python library v3.7.0 updates usage APIs and documentation, and includes a fix for repeated numpy checks in embeddings.

09-02RTNweb app

ChatGPT web app frontend deployed with new internal feature flag and several W3C, OpenAI, and help documentation related endpoint strings.

09-01RTNgithub

OpenAI's Codex rust-v0.152.1 release fixes an issue where Guardian approval reviews were not respecting Node REPL policies from model metadata.

09-01URGENThacker news

Apple has reportedly found 'shocking evidence' on an ex-employee's MacBook related to the OpenAI lawsuit, suggesting potential new developments in the legal battle.

09-01URGENTgithub

OpenAI Codex published a new pre-release, rust-v0.153.0-alpha.4.

09-01RTNweb app

New frontend deploy to ChatGPT web app includes an 'internal' feature flag and new endpoint strings, possibly for internal tools or SDKs.

09-01URGENTgithub

OpenAI's Codex project has published a new alpha pre-release, rust-v0.153.0-alpha.1, for its Rust implementation. ✓ beat the announcement by 15.0h

09-01RTNgithub

OpenAI Codex's `rust-v0.152.0` release introduces Vim search support in drafts, rate-limit banners with usage actions, and credential-refresh progress for Amazon Bedrock.

09-01RTNgithub

OpenAI's Codex project published a new pre-release version: rust-v0.152.0-alpha.7.2.

08-31RTNgithub

OpenAI's Codex project released a new pre-release version: rust-v0.152.0-alpha.7.

08-31RTNweb app

A new feature flag, `generate_trending_suggestions`, was added to the ChatGPT web app, indicating potential upcoming 'trending suggestions' for conversations.

08-31FLASHdocs

OpenAI's changelog indicates the deprecation of existing Whisper and GPT-4o transcription models, with a shutdown set for February 26, 2027, in favor of new 'gpt-live-transcribe' or 'gpt-transcribe' models. Additionally, the Assistants API shut down on August 26, 2026, replaced by Responses and Conversations APIs.

08-31RTNweb app

New strings in ChatGPT web app for a trending suggestions feature flag and various OpenAI help and policy endpoints.

08-31URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.152.0-alpha.6. ✓ beat the announcement by 10.9h

08-31URGENTgithub

OpenAI's Codex project has published a new pre-release version: rust-v0.152.0-alpha.5. ✓ beat the announcement by 11.8h

09-06RTNgithub

A new nightly pre-release, v0.60.0-nightly.20260906.g85aca163f, has been published for the Google Gemini CLI, indicating ongoing development.

09-05RTNgithub

Gemini CLI pre-release v0.60.0-nightly.20260905.g85aca163f includes fixes for extension consent prompts, workspace path checks, and config file permissions.

09-04RTNgithub

Gemini CLI releases v0.60.0-nightly.20260904.g87a9c71d5 with security fixes and extension hardening.

09-03FLASHmodel api

A new model ID, lyria-3.5, has appeared in Google Gemini's API, signaling a potential upcoming release.

09-02FLASHhacker news

An HN post titled "Gemini 3.8 Flash" is gaining significant velocity, potentially indicating a new model release or update.

09-02FLASHhacker news

An HN story titled "Gemini 3.8 Flash and 3.8 Flash Cyber" is gaining significant velocity, indicating a potential announcement or release from Google related to new Gemini models.

09-02FLASHmodel api

A new model ID, 'gemini-flash-latest-high-res-exp', appeared in Google Gemini's API, suggesting a potential high-resolution flash model is in development or testing.

09-02FLASHmodel api

OpenRouter's API has a new model id: google/gemini-3.8-flash:batch.

09-02FLASHmodel api

OpenRouter and Google Gemini APIs show a new model ID, google/gemini-3.8-flash, suggesting an upcoming release.

09-02RTNgithub

Google Gemini CLI published a new pre-release v0.59.0-nightly.20260902.g4963a4456 with an improved destination validation and connection routing in web fetch utilities.

09-01RTNgithub

Gemini CLI v0.58.0 released with fixes for symlink evaluation, Docker isolation on macOS, cancellation errors, and safety checker configuration.

09-01RTNgithub

Google Gemini CLI published a pre-release v0.59.0-preview.0 with fixes for SSRF and workspace trust.

08-31RTNgithub

A new nightly pre-release, v0.59.0-nightly.20260831.g0bd1d4397, has been published for the Google Gemini CLI.

09-04FLASHlmarena

A new Grok model, `grok-4.6-high-public-webdev`, has appeared on the LMArena leaderboard, suggesting a potential future release or testing of a web development-focused model.

09-04FLASHmodel api

OpenRouter API shows a new model ID: x-ai/grok-4.3:batch, suggesting a new Grok version or batch inference capability from xAI.

09-03URGENThacker news

High velocity on Hacker News for 'Grok Outage'.

08-31FLASHhuggingface

Nvidia has published a new HuggingFace repository: nvidia/DeepSeek-V4-Flash-0731-NVFP4. This could indicate an unannounced collaboration or a new model release. 'Flash' in the name suggests it could be a smaller, faster inference optimized model.

08-31FLASHhuggingface

DeepSeek-AI created a new HuggingFace repository named 'DeepSeek-V4-Flash-Vision-Exp', potentially indicating a new experimental vision model release.

09-02URGENThacker news

Mistral's default policy now includes training on user input, excluding the enterprise tier, as evidenced by a fast-rising Hacker News story.

09-05FLASHmodel api

A new model ID, qwen/qwen3.8-max-0902, has appeared in OpenRouter's API, suggesting an upcoming release.

09-04FLASHhuggingface

Nvidia has pushed a new HuggingFace repository named 'nvidia/Qwen3.8-Flash-Next-NVFP4', which appears to be a new model or model variant.

09-03FLASHhuggingface

Qwen pushed a new HuggingFace repository named Qwen/Qwen-Drive-1.0-4B, potentially indicating a new model release.

A paper titled "LLMs as a Cognitive Virus" is gaining rapid traction on Hacker News, suggesting a high level of community interest in the conceptualization of LLMs as a cognitive phenomenon.

OpenLake's blog post claims they beat MLPerf in storage for KV offload and LLM training, gaining significant traction on Hacker News.

A new model, 'saiph-ooo3', has appeared on the LMArena leaderboard, potentially indicating an upcoming release.

A new anonymous model, vega-v9e1, has appeared on the LMArena leaderboard, suggesting a stealth release.

An article about Corporate America's increasing reliance on open-source AI is gaining traction on Hacker News.

LMArena has added a new anonymous model, "eren-v2-2mgj," to its leaderboard, likely indicating an unannounced stealth model in testing.

A new NVIDIA model, nemotron-3.5-content-safety, appeared in OpenRouter's API, suggesting an upcoming release focused on content safety.

A new, anonymously named model 'muse-spark-1.3-xhigh-webdev' has appeared on the LMArena leaderboard, suggesting a stealth model currently in testing.

LMArena has added a new anonymous model, "sirius-thsw-i2v", to its leaderboard, likely indicating an unannounced model being tested.

Microsoft pushed new HuggingFace repositories for VibeVoice-ASR-Streaming models, specifically 1.5B and 7B variants.

A new, anonymously-coded model, 'skimaki-kwc6', has appeared on the LMArena leaderboard, potentially indicating a stealth test of an unreleased model.

Nvidia released two new 1.7B-Base language models to HuggingFace: SDLLM-Duo and SDLLM-EsoLM.

A new, anonymous model, 'kiana-8mgg', has appeared on the LMArena leaderboard, likely indicating a stealth release or unannounced testing.

A new, anonymously coded model, 'august26-chatbot1-fmme', appeared on the LMArena leaderboard, suggesting a stealth release.

Nvidia created a new HuggingFace repository for a model named Muse-Glimmer-30B-NVFP4.

An HN post details a small transformer trained in 1.5 hours outperforming many LLMs, gaining significant velocity.

Two new anonymous models, "farris-ees5-webdev" and "phoenix-qcs2-webdev," appeared on the LMArena leaderboard, alongside "glm-5.3-flash-webdev."