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

10-11FLASHlmarena

Two new Claude Opus 5.5-max-vertex-agent models, one labeled as a 'contender,' appeared on the LMArena leaderboard, suggesting unannounced model testing.

10-11RTNhacker news

An article discussing Anthropic banning 'cruelty' to Claude is gaining traction on Hacker News, though the specifics of what it protects are unclear.

10-10URGENTbinary

Anthropic's Claude Code binary for Darwin ARM64 was updated, showing new strings related to internal feature flags, several API endpoints, and mentions of 'Managed Agents' in beta, self-hosted environments, and debugging tasks on GitHub.

10-09FLASHmodel_catalog

27 new model IDs added to the LiteLLM catalog, including several 'gpt-5' and 'gpt-6' variants, new Claude models, and Gemini 3.7/3.8 flash models under github_copilot. ✓ beat the announcement by 11.0h

10-09RTNsitemap

Anthropic added 24 new sitemap URLs across multiple languages, suggesting new documentation for 'managed-agents', 'session-threads', and 'workflow-runs'.

10-09RTNweb app

Anthropic's Claude platform console deployed new frontend code, including a new billing API endpoint related to Stripe region and an invalid relative URL.

10-09FLASHmodel_catalog

A new model ID, 'github_copilot/claude-haiku-5.5', appeared in the LiteLLM model catalog. ✓ beat the announcement by 15.2h

10-09URGENTgithub_releases+pages

Claude Code v2.1.296 adds `code` key to gateway policies, `autoCompactWindow` for subagents, `CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL` for consistent model use, and an `allow_large` option to the Read tool to handle larger files.

10-09RTNweb app

Anthropic's Claude platform console deployed new frontend code, including new API endpoints related to account management and Stripe billing.

10-09URGENTdocs

Anthropic's API changelog shows new beta features for Claude Managed Agents, including dynamic workflows and updated `multiagent` field, along with changes to error handling and thinking block display.

10-09RTNgithub

Anthropic's Python SDK v1.13.0 adds analytics metrics for Chat and Cowork, and new features for Managed Agents including workflows and multiagent configuration.

10-09FLASHpypi

Anthropic's latest SDK includes new `claude-opus-4-8` and `claude-haiku-4-5` model IDs, alongside beta features for messages streaming, refusal fallback middleware, and environments for workflow management. It also adds new configuration options for agents and workflow runs.

10-09URGENTnpm

Anthropic's SDK updated with new string literals related to tool use, chat prompts, and a `myContext` request option, suggesting enhanced function calling capabilities.

10-09RTNweb app

Anthropic's Claude platform deployed new frontend strings, including an internal tier feature flag and new API endpoints related to Stripe billing and OAuth redirects.

10-09URGENTbinary

Anthropic's Claude Code binary update hints at a new `claude-haiku-3-55` model, internal error handling improvements, and potential long context beta features.

10-09RTNsitemap

Anthropic quietly published 22 new sitemap pages related to 'managed agents', 'event deltas', and 'session observability' across multiple languages.

10-09RTNweb app

Anthropic's Claude platform deployed new frontend strings, including a feature flag for "internal_limited_retention" and new hCaptcha endpoints.

10-08FLASHlmarena

LMArena has added 8 new models to its leaderboard, including a stealth model codenamed 'saga-decima-0kzi' and several named models like 'claude-fable-5-v2' and 'gemini-3.5-flash-agent-high-public'.

10-08RTNgithub_releases+pages

Claude Code v2.1.295 introduces new features like `onFailure: "block"` for hooks, Program Status Protocol support, and enhanced model routing for the Claude apps gateway, alongside various fixes.

10-08FLASHmodel_catalog

A new batch inference model ID, 'openrouter/anthropic/claude-haiku-5.5:batch', has been added to the LiteLLM model catalog. ✓ beat the announcement by 40.1h

10-08URGENTgithub_org

Anthropic made a new public repository named 'oss-scanner', suggesting a tool for open-source software scanning.

10-08RTNbinary

Anthropic's Claude Code for darwin-arm64 added a new endpoint string: claude.ai/code/artifact/.

10-08URGENTgithub_org

Anthropic made a new public GitHub repo called 'snix', which mirrors git.snix.dev/snix/snix.

10-08FLASHlmarena

26 new models, including several stealth codenames (fable-5.1, hy4-preview, inkling, opus-5, sonnet-5), appeared on the LMArena leaderboard, alongside unreleased versions of Claude, Gemini, GPT, Grok, and others.

10-08RTNweb app

Anthropic's Claude platform console deployed new frontend code including an "internal_tier" feature flag and endpoint.

10-08RTNgithub_releases+pages

Claude Code (anthropics/claude-code) released v2.1.294, improving `prompt` and `agent` hook instruction handling to prevent early stopping and correctly block commands.

10-08FLASHpypi

Anthropic's 1.12.1 update includes references to new models, 'claude-opus-4-8' and 'claude-haiku-4-5', alongside new beta features for messages and environments.

10-08URGENTnpm

Anthropic's SDK update includes new examples for tool use and message formats, suggesting improved function calling capabilities.

10-08FLASHsitemap

Anthropic has quietly published new documentation pages referencing a "Claude Haiku 5.5" model, along with new features for managed agents and self-hosted sandboxes. ✓ beat the announcement by 13.7h

10-07RTNweb app

Anthropic's Claude.com console is adding hCaptcha feature flags for various login, payment, and SSO flows, indicating enhanced bot protection.

10-07FLASHmodel_catalog

LiteLLM's model catalog shows a new `openrouter/anthropic/claude-haiku-5.5` model id. ✓ beat the announcement by 18.0h

10-07FLASHmodel api

OpenRouter API shows a new Anthropic Claude Haiku 5.5 batch model. ✓ beat the announcement by 19.0h

10-07URGENThacker news

Meta and Microsoft are reportedly limiting employee use of Claude AI tools due to security and data privacy concerns.

10-07FLASHmodel_catalog

17 new model IDs for 'claude-haiku-5-5' appeared in the LiteLLM model catalog across various cloud providers, suggesting an upcoming Anthropic Claude Haiku update or release. ✓ beat the announcement by 20.1h

10-07FLASHmodel api

OpenRouter API shows new model ID: anthropic/claude-haiku-5.5, suggesting an upcoming Anthropic release. ✓ beat the announcement by 20.1h

10-07URGENTbinary

New `claude-haiku-5-5` model ID and several internal feature flags, endpoints, and sentences detected in an Anthropic Claude Code binary update. ✓ beat the announcement by 20.6h

10-07FLASHgithub_releases+hn+npm+pages+pypi+sitemap

Anthropic's SDK updated to 0.132.0, with a new model ID `claude-haiku-5-5` and several new beta toolset features. Evidence is strong across multiple sources. ✓ beat the announcement by 0.0h

10-07RTNweb app

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

10-07URGENTbinary

New strings in the Claude Code binary suggest a dropped 'claude-dropped-9' model, an internal mid-conversation system beta, and new API endpoints related to frames, plugin ratings, and sessions.

10-06RTNdocs

Anthropic's Models API now reports a new `capabilities.thinking.types.disabled` field, indicating if a model supports turning off 'thinking' with `thinking: {type: "disabled"}`.

10-06RTNweb app

Anthropic added a new API endpoint related to Stripe billing and region information on their Claude platform.

10-06URGENTgithub_releases+npm+pages

Anthropic's claude-code released v2.1.292, adding an `effort` parameter for sub-agents, prompt caching, and various fixes and security updates.

10-06RTNbinary

Anthropic's `claude-code-darwin-arm64` binary update hints at new internal denial features, local development capabilities, and improved shell environment handling, potentially for a coding agent.

10-06RTNweb app

Anthropic's Claude platform deployed new frontend code including a new endpoint related to Stripe billing region information.

10-06RTNgithub_releases+pages

Claude Code released v2.1.291, fixing regressions related to dropped answers in cloud sessions and lost messages.

10-06RTNbinary

New strings in the Claude Code binary suggest internal features, escalation paths, and integrations with various third-party services like GitHub, Amplitude, Asana, and Slack.

10-06URGENTgithub_releases+pages

Claude Code v2.1.290 adds `serverToolUses` to mod hooks, `agentId` to `tool.check` events, new `managed-agents-onboard` commands, and various fixes for long sessions, tool calls, and background tasks.

10-05RTNweb app

Anthropic's Claude console has new frontend strings hinting at a 'remember SSO email' feature and a new user access endpoint.

10-05RTNdocs

Anthropic added a 'line' field to their Models API, allowing users to group models without parsing IDs.

10-05RTNweb app

Anthropic's Claude platform console deployed new frontend code, including a new `anthropic-beta` feature flag, a new AWS external API endpoint, and an updated support article URL.

10-11RTNgithub

OpenAI's Python library v3.29.0 adds a new 'Shida' voice, billing error code, vault updates with metadata filters, and beta pagination alignments.

10-11FLASHpypi

OpenAI's latest client update, 3.29.0, reveals a new `gpt-4o-2024-08-06` model ID and multiple new 'realtime', 'responses', 'live', 'forks', and 'sideband' beta connection features, alongside improvements to JSON mode, streaming, and tool definitions.

10-10RTNgithub

OpenAI Codex published a pre-release for its Rust SDK, version 0.163.0-alpha.5.

10-09RTNnpm

The OpenAI SDK v7.32.0 hints at new 'ranker' options, detailed file search tool configurations including chunking, and JSON mode improvements.

10-09RTNgithub

OpenAI's Python library adds agent environment suspension and expiration features in v3.28.0.

10-09FLASHpypi

OpenAI's latest SDK update (3.28.0) includes a new `gpt-4o-2024-08-06` model ID, new `beta.realtime` and `beta.responses` connection features, and mentions of `agent.environment` lifecycle events and JSON mode improvements.

10-09RTNgithub

OpenAI Codex released rust-v0.162.1, addressing TUI crashes with multiline asynchronous questions and startup failures related to server feature settings.

10-09URGENTmodel_catalog

Two new model IDs, 'chatgpt/gpt-rosalind-discovery' and 'gpt-rosalind-discovery', have appeared in the LiteLLM model catalog.

10-09FLASHsitemap

OpenAI's sitemap quietly published 19 new URLs, including a page mentioning "gpt-5-our-best-model-for-work/". ✓ beat the announcement by 15.9h

10-09RTNhacker news

A Hacker News story titled "The super intelligence shit is a humiliation ritual for OpenAI" is gaining significant velocity, suggesting community discussion around OpenAI's AI safety or AGI claims.

10-09RTNgithub

OpenAI Python library v3.27.0 adds 'prewarmed hosted environments' API feature.

10-09FLASHpypi

OpenAI 3.27.0 adds a new `gpt-4o-2024-08-06` model ID and hints at new `realtime` and `responses` beta features, alongside various client and streaming improvements.

10-09RTNgithub

OpenAI Codex published a pre-release for its Rust library, version 0.163.0-alpha.4.

10-09URGENTnpm

OpenAI's latest SDK update (7.31.0) introduces new file search tool configurations including chunking options and a 'ranker' parameter. It also clarifies JSON mode usage.

10-09URGENTgithub

OpenAI's Codex project has published a new pre-release version: rust-v0.163.0-alpha.2. ✓ beat the announcement by 32.7h

10-09RTNhacker news

A Hacker News story about OpenAI, the Partition Principle, and Mathematics is gaining significant traction.

10-09RTNweb app

A new internal endpoint `chatgpt.invalid` was found in a ChatGPT web app frontend deploy, likely for testing or internal routing.

10-08RTNgithub

OpenAI Codex published a new pre-release version `rust-v0.163.0-alpha.1` for its Rust SDK, indicating ongoing development.

10-08RTNgithub

OpenAI's Codex Rust SDK v0.162.0 adds new features like managed Git worktrees, pinned tasks, improved transcript navigation, clickable URLs, and configurable web access for custom model providers, along with various bug fixes.

10-08RTNhacker news

HN velocity indicates news about OpenAI's annualised revenues being $20B less than previously signalled is climbing fast.

10-08FLASHdocs

OpenAI's changelog indicates the addition of a new 'Ultrafast mode' for 'gpt-6.1-sol' in the Responses API, available to all users with global processing and US/EU data residency. ✓ beat the announcement by 39.9h

10-08RTNhacker news

OpenAI has withdrawn three math papers from its GitHub repository, signaling a potential shift or re-evaluation of their published research in this area.

10-08RTNgithub

OpenAI's Python library received a patch to fix custom voice creation parameters.

10-08FLASHpypi

OpenAI's latest SDK update (3.26.1) includes a new `gpt-4o-2024-08-06` model ID, alongside hints for 'realtime' and 'responses' beta features, and updates to streaming and JSON mode.

10-08RTNnpm

OpenAI's latest npm update (7.30.1) introduces a 'ranker' parameter for Assistants, file search chunking options (overlap, max tokens), and clarifies JSON mode usage.

10-08RTNgithub

OpenAI's Codex project published a new pre-release version: rust-v0.162.0-alpha.17.2, signaling continued development in Rust.

10-08RTNgithub

OpenAI's Codex project released a new pre-release version: rust-v0.162.0-alpha.18.1.

10-08URGENTgithub

OpenAI's Codex project has published a new pre-release, `rust-v0.162.0-alpha.20`. ✓ beat the announcement by 7.8h

10-08RTNgithub

OpenAI's Codex project released a new pre-release version: rust-v0.162.0-alpha.17.1.

10-07FLASHhn+sitemap

A new page titled "GPT-6 and Intelligent UI for everyone" appeared on OpenAI's sitemap and is gaining traction on Hacker News, suggesting an upcoming announcement for a new model or feature. ✓ beat the announcement by 0.0h

10-07FLASHgithub

OpenAI's Codex rust-v0.161.0 release defaults to GPT-6.1 Sol, adds multi-agent V2 to Amazon Bedrock, and introduces Daybreak opt-in. ✓ beat the announcement by 0.0h

10-07RTNgithub

OpenAI Codex published a new pre-release, `rust-v0.162.0-alpha.18`, for its Rust implementation.

10-06URGENThacker news

An OpenAI blog post titled "Sharing AI Progress in Mathematics" is trending fast on Hacker News. ✓ beat the announcement by 1.7h

10-06RTNhacker news

An OpenAI GitHub repo titled "Mathematical manuscripts and supporting proof artifacts produced by OpenAI" is gaining traction on Hacker News.

10-06URGENTgithub_org

OpenAI made a new public GitHub repository called 'math'. ✓ beat the announcement by 0.0h

10-06URGENTgithub

OpenAI's Codex project has published a new pre-release, rust-v0.162.0-alpha.17. ✓ beat the announcement by 18.1h

10-06RTNgithub

OpenAI Python library v3.26.0 adds standalone Decisions support.

10-06FLASHpypi

OpenAI's latest SDK update, 3.26.0, reveals a new `gpt-4o-2024-08-06` model ID and several `beta.realtime` and `beta.responses` connections, possibly hinting at real-time streaming or interactive features.

10-06URGENTnpm

OpenAI's latest npm package (7.30.0) introduces an `agents.toolStages` feature flag and new strings related to tool definitions (code interpreter, file search), chunking/overlap for file search, and JSON mode for model messages.

10-06RTNgithub

OpenAI's Python library v3.25.0 introduces new agent capabilities, including observing local tool failures, running local tools during streamed session creation, and adding agent turn items to the API.

10-06FLASHpypi

OpenAI's latest SDK hints at a new `gpt-4o-2024-08-06` model ID and introduces `realtime` and `responses` beta connections, potentially for live interactions or streaming responses. ✓ beat the announcement by 0.0h

10-06FLASHmodel_catalog

3 new model IDs `gpt-6.1-sol` were detected across Azure's APAC, EU, and US regions in the LiteLLM model catalog. ✓ beat the announcement by 1.1h

10-06RTNgithub

OpenAI Codex published a new pre-release version for its Rust integration: `rust-v0.161.0-alpha.13.1`.

10-06FLASHmodel_catalog

Four new model IDs, including `gpt-6-astra`, `gpt-6-luna`, `gpt-6-sol`, and `gpt-6.1-sol`, have appeared in the LiteLLM model catalog under the chatgpt family. ✓ beat the announcement by 13.9h

10-05RTNgithub

OpenAI's Codex project published a new pre-release: rust-v0.162.0-alpha.16, indicating ongoing development, possibly in their Rust implementation.

10-05RTNgithub

OpenAI's Codex Rust SDK release 0.160.1 fixes environment variable preservation for remote stdio MCP servers, specifically for Windows executors on Unix hosts.

10-05URGENThacker news

An article about "rogue" OpenAI agent activities being found on Wikimedia projects is gaining traction on Hacker News, suggesting potential unauthorized AI behavior or testing on public platforms.

10-05URGENTgithub

OpenAI's Codex project has published a new pre-release version: rust-v0.162.0-alpha.15. ✓ beat the announcement by 27.8h

10-05URGENTgithub

OpenAI's Codex project released a new pre-release version: rust-v0.162.0-alpha.14, indicating active development. ✓ beat the announcement by 38.6h

10-10RTNgithub

Google Gemini CLI published a new pre-release with bug fixes and performance improvements, including better error handling, credential management, and file reading optimizations.

10-09RTNgithub

Google Gemini CLI published a pre-release v0.64.0-preview.1, which includes a patch fix.

10-09URGENTgithub_org

Google Gemini made a new public repository called 'hermes-gemini-api-plugin'.

10-08RTNgithub

Gemini CLI pre-release v0.65.0-nightly.20261008.g44d764ee5 adds support for custom OTLP headers in telemetry configuration.

10-07RTNgithub

Gemini CLI releases v0.65.0-nightly.20261007.gef59c532f with fixes for workspace settings, tool response turns, session history, OAuth, and terminal behavior.

10-07FLASHmodel_catalog

Four new model IDs, including Mistral large-4 variants and a new Gemini Nano version, have been added to the LiteLLM model catalog.

10-06RTNgithub

Gemini CLI releases v0.64.0-preview.0 with V1 to V2 settings migration, usage tracking, and various CLI/core fixes, alongside a stable v0.63.0.

10-06FLASHmodel_catalog

Two new model IDs, 'gemini-nano-banana-2.1' and 'vertex_ai/gemini-nano-banana-2.1', have appeared in the LiteLLM model catalog, suggesting a new Gemini Nano version on Vertex AI.

10-06FLASHmodel_catalog

Three new model IDs, including "azure/model-router," "azure/whisper," and "gemini/gemini-nano-banana-2.1," have been added to the LiteLLM model catalog, indicating potential new offerings or integrations from Azure and Google.

10-06FLASHmodel api

OpenRouter's API added a new model ID: google/gemini-nano-banana-2.1, suggesting a new Gemini Nano version may be coming soon.

10-06FLASHhuggingface

Google pushed a new HuggingFace repository for 'embeddinggemma-2'.

10-06FLASHmodel api

A new model ID, gemini-nano-banana-2.1, has appeared in Google Gemini's API, suggesting an unannounced model.

10-06RTNgithub

Gemini CLI published a new nightly pre-release, v0.64.0-nightly.20261006.gfb972b2f8.

10-05URGENTgithub

google-gemini/gemini-cli published a pre-release: v0.64.0-nightly.20261005.gfb972b2f8

10-06FLASHmodel_catalog

LiteLLM's model catalog now includes 'azure_ai/grok-4.7', suggesting a potential Grok integration on Azure AI.

10-09FLASHlmarena

LMArena's leaderboard has added 5 new models, including two anonymous "mimo-v2-6-pro-public-agent" entries and named models like "mistral-large-4-agent" and "deepseek-v4-pro-thinking-first-party."

10-09FLASHhuggingface

Mistral AI pushed two new HuggingFace repositories: LIDstral-Arabic and Voxtral-Mini-4B-Realtime-Arabic, suggesting new Arabic language capabilities.

10-07FLASHlmarena

LMArena added three new models to its leaderboard, including two anonymous models, ajax-20260517 and mimo-v2-6-pro-public-webdev, and mistral-large-4.

10-06FLASHhacker news

A fast-rising Hacker News story titled "Mistral Large 4" suggests a potential new model from Mistral AI.

10-06FLASHhacker news

High velocity HN signal for a new Mistral model, 'Mistral Large 4: "Le Chonk"'.

10-06FLASHmodel api

A new model ID, mistralai/mistral-large-4-0, has appeared in OpenRouter's API, indicating a potential unannounced Mistral AI model.

10-06FLASHhacker news

A Hacker News story about "Mistral Large 4" is gaining significant velocity, potentially indicating a new model release.

10-10FLASHmodel_catalog

A new model ID, databricks/databricks-openjev-qwen35-4b, has been added to the LiteLLM model catalog, suggesting a new Databricks offering based on Qwen3.5-4B.

10-09FLASHhuggingface

Qwen pushed a new HuggingFace repository for an image model: Qwen/Qwen-Image-2.1-Turbo.

10-07URGENTdocs

Cursor is launching remote control for local agents, allowing users to monitor and interact with agents running on their computers from the Cursor iOS app.

A new model ID, databricks/ai_decide, has been added to the LiteLLM model catalog.

Microsoft pushed new HuggingFace repositories for "AesCode-32B" and "AesCode-8B", likely new model weights.

5 new model IDs (scaledown/classify, compress, decisions, extract, summarize) detected in LiteLLM's model catalog, suggesting new specialized model offerings.

A new model ID, amazon.nova-premier-v1:0, was detected in the litellm model catalog, suggesting a potential new Amazon model.

Two new Cohere embedding models, Cohere-Embed-V5-Fast and Cohere-Embed-V5-Pro, have been added to the Azure AI model catalog.

Nvidia has pushed three new 'agile_one' object detection models to HuggingFace, potentially for robotics or industrial automation tasks.

Microsoft-Decision-1 is a new model ID appearing in the Azure AI model catalog, suggesting a new decision-making model.

A new model ID, 'twelvelabs.pegasus-1-5-v1:0', has appeared in the LiteLLM model catalog.

Nvidia has pushed four new repositories to HuggingFace under the `agile_one_s` naming convention, potentially indicating new models or datasets related to robotics or agile movements.

Two new model IDs, global.twelvelabs.pegasus-1-5-v1:0 and us.twelvelabs.pegasus-1-5-v1:0, have been added to the LiteLLM model catalog, suggesting a new Twelve Labs model integration.

Two new model IDs, 'openrouter/inclusionai/ling-3.0-flash-sante' and 'openrouter/stepfun/step-5-preview', have been added to the LiteLLM model catalog.

A new model, ernie-5.0-preview-1203, has appeared on the LMArena leaderboard, potentially indicating an upcoming release.

LiteLLM's model catalog now includes 'in.moonshotai.kimi-k3'.

Nvidia created a new HuggingFace repository named 'nvidia/so101-reach'.

A new, anonymously listed model called 'mimo-v2-6-flash-webdev' has appeared on the LMArena leaderboard, potentially a stealth release.

A new anonymous model, "river-route-29j0", has appeared on the LMArena leaderboard, potentially a stealth release being tested.

Three new model IDs (amazon.nova-2-5-sonic, global.zai.glm-5.3, us.zai.glm-5.3) added to the LiteLLM model catalog, indicating potential new offerings from Amazon and ZAI.