vLLM + Megatron add Day-0 RL support and long-agent serving for Kimi K3.1
vllm_project · x · 2026-11-05
The vLLM project, working with inferact, detailed vLLM + Megatron support for Kimi K3.1 with Day-0 RL. The work covers training-inference consistency for Kimi Delta Attention plus the cache and state management needed to serve long-running agents — from serving the model to continuing its training on the open stack.
A related post reports K3.1 Pro scoring 61.2% on the internal Kimi Code Bench 3.0 Long-horizon (up from K3's 55.95%), covering multi-step execution, tool use and end-to-end engineering, both measured with the Claude Code harness at maximum reasoning effort.
Related event: vLLM Ships Day-0 Support for Kimi K3.1 Across Training and Serving(5 posts)→
More from Infra
- Deployment link shared for Mistral Large 4 1T endpoint at $10/hr quantized — ngxson · 2026-11-05
- Mistral Large 4 1T hits HF Inference Endpoints at $10/hr on 4x A100 — ngxson · 2026-11-05
- vLLM ships Day-0 support for Kimi K3.1: 61.2% on long-horizon coding bench, RL-ready stack — vllm_project · 2026-11-05
- NVIDIA accelerates structural simulation in Abaqus with CUDA-X — NVIDIAAI · 2026-11-05
- SpecAgent: multi-session Claude Code pipeline with measured 3x context cost savings — AdOtherwise9458 · 2026-11-05
- SoftBank cloud arm IDC Frontier hit by ransomware in 'seven minutes,' taking ~500 organizations offline in Japan — ChuckDBrooks · 2026-11-05