AI Briefing#tool-use
The latest AI and developer news about #tool-use, with the original source and a short summary.
The latest AI and developer news about #tool-use, with the original source and a short summary.
The model succeeds Gemini 3.8 Flash, handling longer engineering tasks and producing better-researched documents in fewer agent steps.
The open-source model achieves SOTA tool-use performance for its size, scoring 45.0 on average compared to 38.8 for Qwen3.5-2B-Think.
Claude Fable 5.1 ranks #3 on SWE-Bench Pro V2 and #2 on MCP Atlas, but its low Remote Labor Index score (17.92%) indicates it requires human verification for autonomous professional tasks.
The open-source model uses a Gemma 4 backbone and Whisper encoder to handle speech, reasoning, and tool calls, outperforming other open-source systems on voice benchmarks.
The new benchmark evaluates 507 stateful business workflows by verifying final database states rather than tool call success, revealing that 67% of agent failures occur despite clean termination.
The new DTOC framework allows agents to dynamically hide or unhide tool outputs, improving solve rates and reducing token costs on DeepSWE tasks.
GLM 5.2 lowered the Tool-call failure rate by 21.2% through combined training using RFT on real user session data and Hint-guided Self-distillation.
The MCP server proofcut for Claude Code has released transcription-based automatic editing and self-verification rendering features.
The open-source library tool-prune has been released, reducing prompt tokens by 92% through pre-filtering of tool schemas.
ToolGrad achieves a higher pass rate using an answer-first approach, while small Gemma-3 models demonstrate competitive performance against SoTA models.
A study on coding agents analyzes the preference for grep over semantic navigation (LSP) and identifies the impact of tool output formats on agent performance.
The 'Auto-BenchMax' data synthesis pipeline, tailored to benchmark scoring methods, has been released and improved the MCP-Atlas score by more than double.
It provides guides for building and optimizing agents by leveraging Claude's various capabilities.
Bad Theory Labs has released BTL-3, a 27B open-weight agentic coding model that runs from a single 8.39GB file, under the Apache-2.0 license.
A study has been published that monitors AI agents' tool-use decisions through Mechanistic Interpretability.
A phenomenon called the 'Verifier Tax' has been discovered, in which the success rate of LLM agents declines as the number of task steps increases during safety verification.
When training agentic RL, you must avoid token mismatch issues that arise from re-encoding decoded tokens.
An attack that manipulates Voice AI using audio inaudible to humans has been identified.
It summarizes how to design agent evaluation harnesses that go beyond static benchmarks.
Explicitly stating source boundaries was needed for LLMs to distinguish text usable as grounds for an answer.
Needle has open-sourced a Gemini-based tool-calling model with 26 million parameters.
Thinking Machines released a research preview of Interaction Models for real-time multimodal interaction.
PORTool improved both accuracy and tool-call efficiency together through step-wise importance estimation.
GLM-5V-Turbo is an agent model that integrates multimodal perception, tool use, and end-to-end verification.