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

10-31RTNhacker news

Anthropic appears to be blocking Claude from being used for reverse-engineering video games.

10-31FLASHhacker news

An AI startup reportedly hacked Anthropic, gaining access to employee details and internal messages.

10-31FLASHmodel_catalog

Four new model IDs, including 'claude-sonnet-5' and 'gpt-5.2', appeared in the LiteLLM model catalog via hostinger_router.

10-31URGENTbinary

Anthropic's `claude-code-darwin-arm64` binary update includes a new model_id reference and numerous internal feature flags, potentially hinting at future capabilities. ✓ beat the announcement by 18.1h

10-31FLASHmodel_catalog

A new Anthropic Claude Haiku model, `eu.anthropic.claude-haiku-4-5-20251001-v1:0`, has appeared in the LiteLLM model catalog, suggesting an upcoming release or regional availability.

10-31URGENTgithub_releases+pages

Claude Code v2.1.316 was released, deprecating fast mode for Opus 4.8 and adding several new features and fixes related to app gateway, subagents, and environment variables. ✓ beat the announcement by 23.5h

10-30URGENTsitemap

Anthropic quietly published 19 new API documentation pages related to a 'claude_tag' feature, including managing bundles and integrating with Slack channels and workspaces.

10-30RTNweb app

Anthropic's Claude platform console deployed new frontend code including new endpoints related to Stripe billing, hCaptcha, and challenge redirects, along with experimental theme feature flags.

10-30RTNhacker news

A fast-rising Hacker News story suggests Blackstone believes Anthropic may not receive an investment-grade rating post-IPO.

10-30FLASHhacker news

A VS Code extension named "Claude Code" on the Cursor Marketplace, which claimed to be for Claude, has been identified as containing a backdoor. ✓ beat the announcement by 31.5h

10-30RTNhacker news

An HN story titled "Anthropic Cracking Down on Video Game Rewrites" is gaining velocity, indicating community interest in potential content policy changes.

10-30FLASHhacker news

A high-velocity Hacker News story suggests Anthropic was hacked to distribute a remote access trojan.

10-30FLASHhacker news

A Hacker News story titled "Backdoors in official Claude Code extensions" is gaining rapid traction, indicating potential security concerns with Claude's tooling. ✓ beat the announcement by 36.3h

10-30URGENTbinary

Anthropic's `claude-code-darwin-arm64` binary update hints at new internal model IDs, feature flags for conversation controls, and several new internal API endpoints, including `/v1/code/channels` and `/v1/filestore/fs/readFileMetadata`. ✓ beat the announcement by 42.0h

10-30URGENTgithub_releases+pages

Claude Code (v2.1.314) adds new features including permission blocks for writes, IdP group support for gateway upstreams, Bedrock guardrail tracing, and idle conversation compaction for VS Code and Claude Desktop. Several fixes were also included related to session failures, error handling, image management in Bedrock, and model fallbacks. ✓ beat the announcement by 47.4h

10-30RTNweb app

Anthropic's Claude platform deployed new frontend code, adding a 'tasks_enabled' feature flag and new hCaptcha related endpoints.

10-29RTNweb app

Anthropic's Claude platform added new API endpoints related to user access, organization memory settings, and organization skills, along with hCaptcha assets.

10-29FLASHmodel_catalog

83 new model IDs, including Claude Fable, Haiku, Opus, Sonnet, Gemini Flash and Pro series, Deepseek, GLM, GPT-5.6 and GPT-6, and other models from various providers were added to the LiteLLM model catalog.

10-29URGENThacker news

An HN post suggests Claude Code might be inefficiently burning 15-20% of tokens on a short status line.

10-28RTNweb app

Anthropic's Claude platform deployed new frontend code including references to a challenge redirect API and hCaptcha.

10-28FLASHmodel_catalog

Databricks and ZeroGPU added 34 new model IDs to the LiteLLM catalog, including several unannounced GPT, Gemini, and Claude versions, as well as new Deepseek, Grok, and Llama models.

10-28URGENTbinary

Anthropic's Claude Code binary update reveals new strings related to 'grove_enabled' feature flag, 'Cyber Verification Program' access for models, and internal shell task monitoring, alongside new documentation and status page endpoints. ✓ beat the announcement by 19.4h

10-27RTNgithub_releases+pages

Anthropic's `claude-code` v2.1.312 release adds OIDC user ID claims for identity, improves message sending, fixes several bugs related to file handling and session management, and refines model behavior with unrecognized custom deployment names.

10-27FLASHdocs

Anthropic launched Claude Sonnet 5.5 and Claude Haiku 5.5, lowered Sonnet 5.5 prompt cache read prices, and added new webhook event types and capabilities to their API. ✓ beat the announcement by 2.9h

10-27URGENThacker news

An article claiming Anthropic used illegally obtained books from Stanford professors to train Claude is gaining traction on Hacker News.

10-27RTNhacker news

A Hacker News story about Anthropic's data usage for model training based on chat feedback is gaining rapid traction.

10-27URGENTct_logs

Anthropic added 5 new subdomains, including API staging for Europe and new relay infrastructure.

10-27URGENTbinary

Anthropic's `claude-code-darwin-arm64` binary update hints at Claude Opus 4 and 4.1 model references, new beta file APIs, and internal debugging/telemetry features. ✓ beat the announcement by 23.5h

10-27RTNweb app

Anthropic's Claude.com platform added two new API endpoints related to Stripe billing and region information.

10-26URGENTgithub_releases+pages

Claude Code v2.1.311 release with new `self-hosted-runner` options, improved context handling, and numerous bug fixes for subagents, file operations, and more. ✓ beat the announcement by 26.3h

10-26URGENTweb app

Anthropic's Claude platform console shows new feature flags for code memory admin export, Google SSO auto-redirect, OIDC federation, and a 'managed agents fable advisor'. Also, new endpoints related to AWS and Google Cloud agent platforms. ✓ beat the announcement by 45.9h

10-26FLASHhacker news

Anthropic's S-1 filing is gaining significant traction on Hacker News, indicating strong public interest in their financial details. ✓ beat the announcement by 0.0h

11-01URGENTgithub

OpenAI's Codex project published a new pre-release: rust-v0.167.0-alpha.2.

10-31RTNhacker news

An article about a hypothetical "GPT-5.6 Sol" model beating Factorio on peaceful mode is gaining significant traction on Hacker News.

10-31FLASHmodel_catalog

Two new model IDs, 'chatgpt/gpt-6-astra-law' and 'gpt-6-astra-law', were added to the LiteLLM model catalog.

10-31URGENThacker news

News of OpenAI raising less than half of its $30B target is gaining significant traction on Hacker News.

10-31URGENTnpm

OpenAI's latest SDK hints at 'file_search' and 'code_interpreter' tool types, alongside configurable chunking for file search and a new 'ranker' option.

10-31FLASHnpm

OpenAI's latest npm package introduces a new model ID `gpt-4o-mini-tts-2025-03-20`, a `/v1/realtime/translations` endpoint, and references to `code_interpreter` and `file_search` tools with new `chunk_overlap_tokens` parameters.

10-31URGENTgithub

OpenAI Python API v3.42.0 introduces new experimental agent features like context compaction, session review, tool settings, and time awareness, alongside image editing, organization key management, and WebSocket improvements.

10-31URGENTpypi

OpenAI's Python SDK v3.42.0 hints at experimental agent settings, new `realtime` and `responses` beta connections, and new vector store file batching capabilities.

10-30FLASHmodel_catalog

Two new model IDs, 'chatgpt/gpt-6-rosalind' and 'gpt-6-rosalind', have been added to the model catalog, potentially indicating a future OpenAI release.

10-29FLASHgithub

OpenAI's Codex adds support for GPT-6.1 Terra in its latest rust-v0.166.0 release, alongside new keyboard navigation and improved Markdown rendering.

10-29URGENTgithub

OpenAI's Codex project has released a new pre-release version: rust-v0.166.0-alpha.6.

10-28RTNgithub

OpenAI Python SDK v3.41.0 introduces new API features for Realtime response inputs, tool usage models, agent session filtering by metadata, and improved WebSocket recovery, alongside documentation updates for image models and upcoming model shutdowns.

10-28FLASHpypi

OpenAI's latest SDK update (3.41.0) includes a new `gpt-4o-mini-tts-2025-03-20` model ID, a `/v1/realtime/translations` endpoint, and multiple references to `realtime` and `responses` beta connections, indicating potential real-time streaming and response handling features.

10-28URGENTgithub

OpenAI's Codex project published a new pre-release, `rust-v0.166.0-alpha.5.1`.

10-28FLASHmodel_catalog

12 new model IDs detected across Azure, Fireworks AI, and Moonshot, including a new 'gpt-6.1-terra' model family on Azure.

10-28FLASHmodel_catalog

13 new model IDs, including deepseek-v4.1-flash, gemma-4-31b-it, kimi-k3, and gpt-oss-120b, were added to the LiteLLM model catalog.

10-28URGENTgithub

openai/codex has published a new pre-release, rust-v0.166.0-alpha.5, on their GitHub.

10-28RTNgithub

OpenAI Codex published a pre-release for its Rust SDK, version 0.166.0-alpha.4.

10-27URGENTgithub

OpenAI's Codex project has released a new pre-release version: rust-v0.166.0-alpha.3.

10-27FLASHgithub

OpenAI's Codex adds experimental GPT-6.1 Terra model, built-in Google Cloud Vertex AI provider, and enables memory v2 by default.

10-27FLASHmodel_catalog

A new model ID, bedrock_mantle/openai.gpt-6.1-terra, has been detected in the LiteLLM model catalog.

10-27RTNgithub

OpenAI Python library v3.40.0 adds prompt cache breakpoint TTL options.

10-27URGENTpypi

OpenAI's Python SDK v3.40.0 adds hints for 'realtime' and 'responses' beta features, along with new polling helpers for vector stores and JSON mode guidance.

10-27URGENTnpm

OpenAI's latest npm package update (7.45.1) hints at new 'code_interpreter' and 'file_search' tool types, 'chunk_overlap_tokens' for file search, a 'ranker' option with a 'default_2024_08_21' value, and clearer JSON mode instructions.

10-27URGENTgithub

OpenAI Codex published a pre-release for its Rust-based version, `rust-v0.166.0-alpha.2`.

10-27RTNgithub

OpenAI's Codex project has released a new Rust pre-release, version `0.165.0-alpha.1.2`.

10-27URGENTgithub

OpenAI Codex published a pre-release for its Rust-based version, `rust-v0.166.0-alpha.1`.

10-27FLASHmodel api

OpenRouter's API added new model IDs: openai/gpt-6.1-terra-pro and openai/gpt-6.1-terra-pro:batch.

10-27URGENTgithub

OpenAI's Codex project has released a new pre-release version: rust-v0.165.0-alpha.1.1, indicating active development.

10-26URGENThacker news

An OpenAI blog post about working with academic communities to build scientific understanding is gaining significant traction on Hacker News.

10-26URGENTgithub

OpenAI Codex released a new pre-release version: rust-v0.165.0-alpha.3. This is an alpha release, indicating early development.

10-26FLASHmodel_catalog

Three new model IDs detected, including 'github_copilot/gpt-6.1-terra' and two 'baseten/decision-models' entries.

10-26FLASHmodel_catalog+npm

OpenAI has shipped version 7.45.0 of their npm package, which includes a new model ID `gpt-6.1-terra` and references to new file search chunking options like `chunk_overlap_tokens` and `chunk_size_tokens`, along with `code_interpreter` and `file_search` tool types. There's also mention of a new `ranker` option.

10-26FLASHhacker news

An unannounced model, GPT-6.1 Terra (Max), is gaining traction on Hacker News due to its performance score of 48 on Artificial Analysis.

10-26FLASHgithub_releases+model_catalog+pypi

OpenAI's latest v3.39.0 release hints at a new model, `gpt-6.1-terra`, along with new real-time connection features and enhanced vector store file handling.

10-26FLASHhn+sitemap

OpenAI quietly published two new pages to its sitemap, including one introducing "GPT-6.1 Terra," which is also gaining traction on Hacker News.

10-26FLASHapi_models+openapi+pages

OpenRouter API and OpenAI's changelog show a new model, `openai/gpt-6.1-terra`, with a batch version, described as fast, efficient for visual reasoning, precise computer use, and natural writing.

10-26RTNgithub

OpenAI Python library v3.38.0 adds delegation controls, prompt cache lifetimes, and expands Live Decision item and question types.

10-26URGENTpypi

OpenAI 3.38.0 adds new beta connection methods for 'realtime' and 'responses', and expands AssistantTool limits and functionality, including new agent and automation features.

10-26RTNnpm

OpenAI's latest npm update (v7.44.0) hints at new file search chunking options, a new 'ranker' parameter, and confirms existing features like JSON mode and various tool types.

10-31RTNgithub

Google Gemini CLI published a new pre-release `v0.67.0-nightly.20261031.g6814e00d6` with several core and CLI fixes, including better handling of ignored files, SSE event flushing, Windows pathing, concurrent extension installs, symlinked paths, custom theme UI, and model selection preservation.

10-30URGENTgithub

Gemini CLI pre-release v0.67.0-nightly.20261030.gb633c0703 includes a new feature for extension validation and evaluation tests, alongside fixes for multimodal function responses for dotted Gemini 3 models.

10-29FLASHhacker news

A Hacker News post titled "Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5" is gaining significant velocity, indicating a new Gemini model may have topped a benchmark.

10-28FLASHhacker news

An HN post titled "Gemini 4 Carbon" is gaining significant velocity, potentially indicating a new model or iteration from Google.

10-28RTNgithub

Gemini CLI published a new pre-release (v0.67.0-nightly.20261028.gfa5c17365) with several bug fixes, including Android path resolution, line ending normalization, terminal rendering, and history/telemetry handling.

10-27RTNgithub

Gemini CLI v0.66.0 release updates the roadmap to remove 2025 references and adds a banner noting it's not optimized for Gemini 4, alongside numerous UI and core bug fixes.

10-27URGENTgithub

Gemini CLI pre-release v0.67.0-preview.0 adds new 'models' command for discovery and IDE companion bridging, plus various fixes.

10-27RTNgithub

Gemini CLI releases a new nightly pre-release v0.67.0 with several bug fixes, including better error classification for spending caps and improved permission requests.

10-29RTNgithub_org

xAI has made a new public repository called 'customer-education-workshops' which contains public demos for workshops.

10-30FLASHhuggingface

Nvidia has pushed a new HuggingFace repository: nvidia/DeepSeek-V4.1-Flash-1017-NVFP4, potentially indicating a new model or model variant.

10-30FLASHmodel_catalog

Two new model IDs, 'aiand/deepseek-ai/deepseek-v4.1-flash-1017' and 'aiand/meta/muse-spark-open', have appeared in the LiteLLM model catalog.

10-30FLASHmodel_catalog

LiteLLM's model catalog now includes 11 new model IDs, featuring Deepseek, GLM, Kimi, Qwen, and Fireworks AI models.

10-29FLASHmodel_catalog

Litellm's model catalog now lists 26 new model IDs, primarily from DeepSeek, GLM, and Kimi, suggesting broader support for various new, potentially unannounced, models.

10-29FLASHhacker news

Mistral Large 4, codenamed 'Le Chonk', has reportedly been released as open-weight, based on rapidly trending Hacker News activity.

10-29FLASHhuggingface

Mistral AI has pushed two new HuggingFace repositories, `mistralai/Mistral-Large-4-1T-A52B` and `mistralai/Mistral-Large-4-1T-A52B-NVFP4`, potentially indicating new model weights.

10-29FLASHmodel_catalog

6 new Mistral 'Voxtral' model IDs, including mini-3b-2507 and small-24b-2507, were detected in the Bedrock model catalog across ap-northeast-1, ap-southeast-2, and eu-west-2 regions.

10-30FLASHmodel_catalog

7 new model IDs detected in the LiteLLM model catalog, including several dashscope/qwen-audio-tts models and two eu_jev models.

10-26FLASHdocs

Cursor launched an Agent SDK for defining, testing, and deploying AI agents as code, with support for various channels and eval methods.

A Hacker News story titled "Malware uses LLMs to rank victims by how much they're worth" is gaining significant traction, indicating a new trend in AI-powered cybercrime.

Deepgram added two new model IDs, deepgram/aura and deepgram/aura-2, to their model catalog.

Microsoft pushed a new HuggingFace repository named `microsoft/UserLM-2.0`.

Two new model IDs, rerank-v5.0-fast and rerank-v5.0-pro, have been added to the model catalog.

Intel's stock dropped 25% due to a bleak outlook, causing a global chip market decline, as reported on Hacker News.

Microsoft created a new HuggingFace repository named 'Mage-VL-Cowork', potentially indicating a new model or project.

LiteLLM model catalog added a new model ID: zai/glm-5v-turbo.

A new model ID, baseten/decision-models/baseten-joey-2b-latest, has been added to the LiteLLM model catalog, suggesting an expansion of available models.

A Hacker News post titled "LLM Voice" is gaining significant traction, suggesting community interest in how LLMs convey tone and style.

6 new Soniox speech-to-text model IDs (stt-rt-preview, stt-rt-preview-v2, stt-rt-v3, stt-rt-v3-preview, stt-rt-v4, stt-rt-v5) detected in the LiteLLM model catalog.

NVIDIA created new HuggingFace repositories for 'NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8' and 'NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-NVFP4', suggesting a new model or model variant is coming.

Three new model IDs (bigmodel/glm-5.3, bigmodel/glm-5.3-flash, bigmodel/glm-5.3-flashx) have been added to the LiteLLM model catalog, suggesting an upcoming release from 'bigmodel'.

LiteLLM has added three new Tencent Seedream image generation models: v5.0-flash, v5.0-lite, and v5.0-pro.

A high-velocity Hacker News story indicates a leak exposing 41,000 Homeland Security workers, including undercover ICE agents.

Nvidia uploaded two new Nemotron 3.5 Transcribe 30B repos to HuggingFace, potentially signaling an upcoming ASR model release.

LiteLLM model catalog added 8 new 'morph' model IDs, including 'morph-glm53-744b', 'morph-glm53flash', 'morph-kimik3', and 'morph-dsv41flash' variants.

Nvidia created a new HuggingFace repository named 'nvidia/Metatron-v1', which could indicate an upcoming model, dataset, or space.

Nvidia created a new HuggingFace repository named 'metatron-browser'.

A 'Survival LLM' project is gaining traction on Hacker News, offering an AI assistant and emergency library that runs locally on a personal computer.

9 new model IDs, including 'refuseless' and 'abliteration' branded models, appeared in the LiteLLM model catalog, suggesting new offerings or finetunes.

A paper titled "The Exponential Time Hypothesis Is False" is gaining rapid attention on Hacker News.

An article about Flock Safety's CEO downplaying public backlash in leaked audio is gaining significant traction on Hacker News.

A new OS-level package firewall has been open-sourced, developed in response to the recent LiteLLM backdoor incident.

Nvidia created a new HuggingFace repository called 'MotionBricks'.

Microsoft pushed a new HuggingFace repository named 'CogACT-v2-RoboDojo', potentially indicating a new model or dataset related to robotics.