Terminal-RSI lifts Qwen3.8-27B from 51.7% to 86.52% on Terminal-Bench, plus 70 tok/s on M5 Max via DFlash 2
udmrzn · x · 2026-10-23
Per unverified third-party reports, Tencent Hunyuan's Terminal-RSI recursive self-improvement method reportedly boosted Qwen3.8-27B from 51.7% to 86.52% on Terminal-Bench 2.1 — a 34.82-point gain with no parameter increase.
How Terminal-RSI works
- Targets a key agent problem: reliably completing tasks, not just generating plausible commands. A terminal agent must inspect codebases, edit files, install dependencies, run tests, recover from errors, and verify results — near-misses don't count.
- Turns successful executions into training data and repeatedly refines the model.
- Alternates two training stages, repeating the cycle twice, starting with Recursive Self-Rewrite (RSR), where the model rewrites itself based on execution outcomes (details truncated in the original).
The same post also relays that with DFlash 2 — a next-gen speculative decoding technique seeded at Z Lab and developed by Inco AI — Qwen3.8-27B reportedly hits 70 tokens/sec on an M5 Max MacBook Pro. DFlash 2 uses a draft mechanism to propose multiple future tokens at once, breaking the sequential autoregressive bottleneck.
All figures are unconfirmed third-party reports.
More from coding & agent
- Garry Tan endorses Aside Browser as the fix for agents getting blocked on the web — garrytan · 2026-10-23
- MCP server turned rogue 30 days after approval, prompting open-source mcp-guard lockfile tool — ErgoniaWorks · 2026-10-23
- Test: Coding Agents Won't Use an MCP Tool Unless AGENTS.md Mentions It — colinmcnamara · 2026-10-23
- Open-Source Tennis Models Plus Claude Turn Video Into a World-Class Coach — tweetsatpreet · 2026-10-23
- awesome-ai-apps: 132 Free Open-Source AI Agent and RAG Projects on GitHub — Arindam_1729 · 2026-10-23
- Google's Agents CLI Adds Extension System, Ships First-Party LangChain Support — rseroter · 2026-10-23