Kimi K3.1 Pro tech blog: stack-wide RL scaling, 65.7% on Terminal-Bench 4.0
natolambert · x · 2026-10-24
Key points from Kimi's K3.1 Pro tech blog:
- Positioning: a post-trained version of Kimi K3 with substantially larger post-training compute; the most capable 3T-class open-weight model with 1M-token context. A 2-3x faster Highspeed variant ships too; weights open by Nov 4.
- Training: to enable large-scale joint RL across hundreds of downstream domains, the team optimized every layer of the stack — RL framework, long-context inference, training efficiency, sandbox infrastructure, rollout scheduling, and a shared environment-quality platform — so each increment of RL compute yields stable, predictable gains.
- Benchmarks: matches or approaches the strongest proprietary models. 65.7% on Terminal-Bench 4.0 vs GPT-6 Astra's 58.2% and Claude Opus 5.5's 66.4%; 61.2% on the internal Kimi Code Bench 3.0 Long-horizon (K3: 55.95%).
- Capabilities: from quick fixes in real codebases to full-project builds, including visual styling.
Related event: Moonshot Releases Kimi K3.1 Pro, Open-Weight Model Rivaling GPT-6(33 posts)→
More from Models
- Kimi 3.1 Pro priced like Argon but scores like DeepSeek V4.1 Flash on meme 'Runescape Bench' — scaling01 · 2026-10-24
- Jon Barron details his workflow: Gemini runs experiments, drafts the paper, then rewrites in his style — jon_barron · 2026-10-24
- Users praise Gem Argon for cultural fluency: a frontier model that roots for you — sebkrier · 2026-10-24
- User says Opus silently downgraded to 4.8 mid-task, then told him to talk to people — natesiggard · 2026-10-24
- Suspicious new GitHub accounts reference Qwen4 in Alibaba T-Head repo PRs — brainExploded99 · 2026-10-24
- Kimi K3.1 Pro takes on the pelican-on-a-bicycle 3D test: beats Opus 5, trails 5.5 — blaind0 · 2026-10-24