DeepSeek Ships V4.1 Flash 1017: Gains on All 11 Benchmarks at 1/16 the Price of Rivals
Prompt Engineering · youtube · 2026-10-18
DeepSeek released V4.1 Flash 1017, dubbed a "minor" update, yet it improved on all 11 benchmarks in the company's own table, with Terminal-Bench 4.0 jumping from 31.2% to 42.3% just five weeks after V4.1 Flash.
- Hits 77.0% on DeepSWE v1.1 at $0.20 per task, the highest score on DeepSeek's cost chart
- Claims parity with OpenAI's GPT 5.6 Sol at roughly 1/16 of Sol's output price
- Same size as V4.1 Flash, further trained with RL on building complete projects
- Live on the DeepSeek API as deepseek-flash at the same price, open weights under MIT license
- Search and Fetch APIs also opened at $0.75 per 1,000 requests
Caveat: all benchmark numbers are DeepSeek's own; independent results are not in yet.
More from Models
- DeepInfra cuts DeepSeek V4.1 Flash price 51% as agent demand surges — gharik · 2026-10-18
- SemiAnalysis: vLLM hits 1.3B tokens per $ TCO, 4x better than B300 aggregated serving — woosuk_k · 2026-10-18
- Nathan Lambert questions Kimi serving Claude tokens directly to users — natolambert · 2026-10-18
- Don't pick your AI based on influencers: frontier models are nearly equal, choose by UI and multimodal needs — techartist_ · 2026-10-18
- Gemini's Redo with Pro vanished a day after launch, frustrating Ultra users — Artistic-Reviewer406 · 2026-10-18
- TypeSafe raises $870M at $7.5B valuation under a month after Jev model launch — dl_weekly · 2026-10-18