DeepSeek V4.1 Flash 1017 tops Vals Index open-weight ranking at just $0.64 per task
scaling01 · x · 2026-10-19
Eval firm Vals AI's latest Vals Index puts DeepSeek V4.1 Flash 1017 (10/17) at #1 among open-weight models with 56.58% accuracy, costing just $0.64 per task at 36m 33s latency.
Key rankings:
- #2 MiMo V2.6 Pro (55.20%, $0.41); #3 GLM 5.3 (53.51% but pricey at $7.25/test)
- The previous DeepSeek V4.1 Flash still holds #5 (51.32%, $0.33), so the new version is a clear step up
- Kimi K3 lands #6 (50.30%); Mistral Large 4 and Qwen 3.8 Max trail under 49%
Combining accuracy and cost, the new DeepSeek stands out on price-performance among open-weight models.
More from Models
- dsh-purge jailbreak plugin for DeepSeek Harness trends on GitHub with 7.3k stars — YuJunZhiXue · 2026-10-19
- 'The Classifier Blocked My Third Try': Devs Are Fed Up With Overzealous Guardrails — fforres · 2026-10-19
- Gemini 4 Argon lands in Google Antigravity for AI Pro and Ultra users — _philschmid · 2026-10-19
- Users notice Gemini's Astra poetry keeps circling themes of nostalgia and impermanence — Tasty-Trip5518 · 2026-10-19
- DeepSeek v4.1 Flash dubbed 'GPT 5.6 Sol at home' in viral AI banter — Dan_Jeffries1 · 2026-10-19
- Robotics Builder: GPT-6 with Controller-Level Function Calling Could Nail 100 Apple Pick-and-Placements by 2026 — wzenus · 2026-10-19