DeepSeek V4.1 Flash fixes 24 of 105 planted bugs for $1.80, near Opus 5's $51.33 run
iPaulis · x · 2026-09-11
Bug Hunt Bench by Pawel Huryn grades frontier coding models on 105 hidden bugs planted in real repos, blind-graded with one prompt per repo. Results: Opus 5 (max) 27 fixes at $51.33, Grok 4.6 (max) 27 at $16.96, DeepSeek V4.1 Flash (max) 24 at just $1.80, GPT-5.6 Luna (xhigh) 23 at $2.50, Opus 5 (high) 21 at $38+.
Takeaway: DeepSeek V4.1 Flash is a genuinely strong everyday coding model at roughly 1/28th the cost of top-tier Opus. Cost spread across the board is 200x while runtime spans under 7x. Qwen 3.8 retests via the official API and more effort tiers are planned.
Related event: Bug Hunt Bench Ranks Frontier Coding Models on 105 Real Bugs(13 posts)→
More from coding & agent
- Dev mixes Claude specs with abliterated GLM to build SS7 threat detector — l4rz · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- GitHub Next's Chopin prototype explores real-time multiplayer agentic planning — max__drake · 2026-10-27
- TMLR paper: enforced thinking doesn't reliably make LLM agents disclose key info to users — SharonYixuanLi · 2026-10-27
- LLM-written Rust library bbtree verified: speed claims hold, memory savings reverse — lmoroney · 2026-10-27
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27