DeepSeek ships V4.1-Flash-1017, now the #1 open model on RSI-Inference Bench
scaling01 · x · 2026-10-22
DeepSeek officially released V4.1-Flash-1017, claiming a Pareto improvement over the previous version: stronger benchmarks, faster task completion, and shorter outputs. It's live on web, app, and API, optimized for DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy.
Day-0 testing on RSI-Inference Bench shows it as the #1 open model, taking #8 overall past GLM-5.3: hidden-test pass rate jumped 35% → 62% (Pass@1 still 0%), with a 1.23× inference speed-up ranking 7th.
More from Models
- DeepSeek code-generated video cost $2.86, but the author says the figure may be off — victormustar · 2026-10-22
- DeepSeek V4.1 Flash renders isometric video purely with code: 4 revisions, $2.86, 549K tokens — victormustar · 2026-10-22
- OpenAI's GPT 6.1 models are coming to Google Cloud and Antigravity — Ars Technica AI · 2026-10-22
- YC F26 startup Latent reads model internals to catch hallucinations at 530x lower cost — ycombinator · 2026-10-22
- Perplexity's JOLT matches GRPO with 6.5x fewer training tokens, JOLT+ wins on all five benchmarks — perplexity_ai · 2026-10-22
- Recursive AI ships Flint, a 40B-A2.5B MoE claimed to match Qwen3.6-27B, weights on Hugging Face — ChengleiSi · 2026-10-22