BTL-3 launches as a 27B open-weight agent model in an 8.39GB file
QuixiAI · x · 2026-07-23
RT of the launch of BTL-3, a 27B open-weight agent model for coding and tool use.
- The full model is compressed into a single 8.39 GB file at under 2.5 bits per parameter.
- It is trained for the agent loop: reason, act, inspect results, recover, continue.
- Reported scores include 95.12% HumanEval pass@1, 88.5% BFCL v4 AST, 95.5% multiple tool calls, and 91.2% tool-call abstention.
- The model supports 262K context and ships in two editions: a high-quality checkpoint for Transformers/vLLM and a standalone GGUF compact version.
More from coding & agent
- Researchers debate whether a code harness, not the base model, is the real unit of comparison — srush_nlp · 2026-07-23
- AI Automated Lead Gen: Targeting New Restaurants Before They Open — eptwts · 2026-07-23
- Open codebases are visible to everyone, but production still needs review — mmitchell_ai · 2026-07-23
- Browser automation stack pairs Browserbase, Stagehand, Weave evals, and Redis memory — wandb · 2026-07-23
- Parallel coding agents hit a human bottleneck: approvals, not compute — casualhermit · 2026-07-23
- Technical Debate: Should Base Transformers Come Equipped with Code Tools — a1zhang · 2026-07-23