VulcanBench VulcanBench
Model results · DeepSeek

DeepSeek V4.1 Flash

Every published VulcanBench measurement of DeepSeek V4.1 Flash (DeepSeek), newest first: combined score, Code quality, runtime, tokens and cost at every reasoning level DeepSeek’s API offers on Frontier v4, graded by deterministic hidden tests on real tasks. Measured as deepseek-flash (DeepSeek-V4.1-Flash-1017 on DeepSeek’s pricing page) through DeepSeek Harness 0.2.1-rc.2, DeepSeek’s own agent, on a metered DeepSeek API key, with its session-log and telemetry uploads off. Its Code quality is judged by Muse Spark 1.3 and GPT-6.1 Sol, a different second judge from most columns, because Grok 4.6’s Cursor account was out of usage. For the earlier DeepSeek V4-Flash (Report 13, August 2026), see its own page.

Suitev4
DeepSeek V4.1 Flash across every effort level
In DeepSeek Harness at Low, High and Max: combined score 88.65, 88.61 and 88.36 (judged n=23 at every level), flat across levels. Tasks passed 21, 20 and 20 of 23; 8.1 to 10.6 minutes and $0.058 to $0.074 a task at DeepSeek’s off-peak list price ($0.068 to $0.101 billed). Best level Low, 22nd of 70 columns on the board. At Low it scores above Claude Opus 5.5 and GPT-6.1 Sol; at High and Max it trails Opus 5.5 and is level with GPT-6.1 Sol, whose columns have a different judge panel.
October 23, 2026 · 23 tasks · 3 effort levels · 69 runs of DeepSeek V4.1 Flash · VulcanBench Frontier v4 · Code quality protocol v3.27
→
Next

How would DeepSeek V4.1 Flash do on your stack?

Public results answer "which model is stronger on this suite." Whether DeepSeek V4.1 Flash belongs in your routing table depends on your languages, your codebases, and your task mix: that gets measured, not guessed. Grading rules are in the methodology.

Optimize your model routing All reports