DeepSeek-V4.1-Flash tops CodeArena open-source web dev leaderboard at ~1/10 Kimi K3 cost
智东西 · wechat · 2026-10-19
DeepSeek-V4.1-Flash-1017 (Max) scored 1670 on Arena's CodeArena:WebDev leaderboard, becoming the top open-source model — beating Kimi K3 (Max) and closed models like Claude Opus 5 and Grok 4.7, just 12 points behind GPT-6 Sol. It also beat GPT-6 Astra on DeepSWE v1.1, with per-task costs down 60% vs the previous version.
Pricing: 2 CNY/M input tokens (cache miss) and 8 CNY/M output at peak, as low as 0.04 CNY with cache hits — roughly 1/10 of Kimi K3's overall cost. The model sits on CodeArena's Pareto frontier: no model is both cheaper and higher-scoring.
Reporters generated a 3D ancient Kaifeng scene and a web racing game in single attempts for 6 CNY total, finding strong spatial and gameplay quality but simplistic character rendering.
Related event: DeepSeek V4.1-Flash-1017: 'Minor Update' Tops Open-Weight Leaderboards(12 posts)→
More from Models
- Meta's Muse Wonder tops DeepSWE coding agent leaderboard at 78.8%, ahead of Claude Opus and GPT-6 — AccBalanced · 2026-10-20
- Alex Atallah's CAP Theorem for Inference Providers: Reliability, Cost, Speed — AccBalanced · 2026-10-20
- lintool's Gaggle Matches SFT'd Qwen With Just a Handful of GPUs on nanochat Backbone — lintool · 2026-10-20
- MiniMax ships 428B MoE model M3.1 Flash at $0.15 per million input tokens — VectorWireAI · 2026-10-20
- Gemini 3.8 Flash missing in Mumbai region, devs weigh Google's data residency churn — Tech_Partner_01 · 2026-10-20
- Report: Liang Wenfeng planned independent DeepSeek compute chips before Huawei persuaded him otherwise — teortaxesTex · 2026-10-20