AI developer tools · What shipped

Based on the website content from the Claude Code GitHub releases page, here is the summary relevant to your user intent about AI developer tools releases, AI coding assistants benchmarks, and machine learning framework updates: Claude Code v2.1.278 introduced changes to auto mode for Claude API and Enterprise users on Bedrock, Vertex, Foundry and gateways, defaulting to server-side classifier billing without classifier overhead charges. Version 2.1.277 added AGENTS.md support allowing Claude Code to read agent configuration files in projects without CLAUDE.md, and version 2.1.275 added a send-now key (ctrl+enter) to interrupt and send queued messages, plus syncing of skills and plugins enabled on claude.ai accounts to terminal sessions. Recent versions include multiple bug fixes and reliability improvements across permission handling, MCP server connections, artifact management, and session resumption features.

GPT-6 Astra set new records on Epoch AI's benchmarks as of September 3, 2026, achieving top performance on the Epoch Capabilities Index (ECI), math, continual learning, and game-puzzle benchmarks following pre-release access by OpenAI. Epoch AI also launched FrontierMath Erdős, featuring 68 unsolved mathematical problems where AI systems must write solutions in Lean; GPT-6 Astra solved 2 of 68 problems (3%), the first model to solve any of them. The benchmarking database now tracks 391 models across 85 benchmarks including GPQA Diamond, MATH Level 5, Mock AIME 2024-2025, and SWE-bench Verified, with evaluation code and detailed model outputs publicly available through log viewers and downloadable data in CSV format via the Epoch AI Python client library.

Open Code Review, released by Alibaba, is an AI-powered code review CLI tool that has been battle-tested serving tens of thousands of developers at scale. According to its benchmark on the AACR-Bench dataset (built from 50 open-source repositories, 200 real pull requests, and 10 programming languages, cross-validated by 80+ senior engineers), Open Code Review achieves significantly higher precision and F1 scores compared to general-purpose agents like Claude Code while consuming only ~1/9 of the tokens. The tool uses a hybrid architecture combining deterministic engineering with an agent for file selection, rule matching, and positioning accuracy, featuring scenario-tuned prompts and toolsets optimized from production data, with integrations available for Claude Code, Cursor, Kimi Code, and other coding platforms.

WSO2 has released WSO2 Agent Manager as open-source general availability software, a platform for centralized governance and identity management of AI agents across different models, frameworks, and deployment environments. The release adds agent identity capabilities, governance controls for Model Context Protocol interactions, and a Kubernetes-native sandboxed runtime. The platform supports multiple frameworks including LangChain and CrewAI, uses OpenTelemetry for tracing, and includes over 40 built-in controls for policies like PII masking and rate limiting, with rule-based and LLM-based evaluation capabilities for monitoring agent behavior over time.