FireFlash Trains 300B Agent on 64 B300 GPUs in Under Five Days

1 reports archived1 independent sourcesupdated 11/12/2026, 20:27:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Researchers released FireFlash, a 300B terminal and coding agent trained via full-parameter reinforcement learning from GLM-5.3-Flash, reaching 46.2% Pass@1 on Terminal-Bench 4.0, on par with Fable 5 and GPT-6 Sol while costing over 30x less per task than Fable 5 at maximum reasoning effort. The entire run used just 64 NVIDIA B300 GPUs for under five days, and the team published the full long-horizon agentic RL recipe.

LATEST/Researchers released FireFlash, a 300B terminal and coding agent trained via full-parameter reinforcement learning from GLM-5.3-Flash, reaching 46.2% Pass@1 on Terminal-Bench 4.0, on par with Fable 5 and GPT-6 Sol while costing over 30x less per task than Fable 5 at maximum reasoning effort. The entire run used just 64 NVIDIA B300 GPUs for under five days, and the team published the full long-horizon agentic RL recipe.

Hourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

Coverage timeline

Total 1 reports · Latest first

机器之心 (微信公众号)T2Nov 12, 20:27·88 pts
FireFlash Trains 300B Agent on 64 B300 GPUs in Under Five Days
- FireFlash hits 46.2% Pass@1 on Terminal-Bench 4.0, matching Fable 5 and GPT-6 Sol at over 30x lower per-task cost
- Training used only 64 NVIDIA B300 GPUs for under five days, with the full long-horizon agentic RL recipe released
- Gains transfer to the DeepSWE v1.1 and SWE-Bench Pro Verified software engineering benchmarks
View Related Reports & Analysis Original Source