DeepSeek Unveils 'V4.1-Flash-1017' with Faster Task Completion and Benchmark Gains
DeepSeek has released 'V4.1-Flash-1017', a minor update reporting shorter task-completion times, shorter outputs, and higher benchmark scores. The model is avai
Chinese artificial intelligence startup DeepSeek has officially unveiled 'V4.1-Flash-1017', a minor update to its existing V4.1-Flash model. DeepSeek reports stronger benchmark results, faster task completion and shorter outputs.
Image source: DeepSeek official release notes (2026-10-17)
1. Key Benchmark Performance Improvements
According to benchmark data released by DeepSeek, V4.1-Flash-1017 achieves notable performance gains across challenging evaluations:
- Humanity's Last Exam (HLE): In DeepSeek's no-tools evaluation, the score rose from 36.8 to 42.6 on the full set and from 39.1 to 44.0 on the text-only subset. These are separate evaluation subsets, not score ranges.
- DeepSWE Benchmark: According to Yonhap Infomax's report of DeepSeek's task-cost chart, V4.1-Flash-1017 achieves a DeepSWE score of 77.0 at roughly $0.20 per task. A lower-cost configuration scores above 72 on DeepSWE at roughly $0.10 per task, while the previous V4.1-Flash scores 74.2 at roughly $0.50 per task.
In DeepSeek's own evaluations, the update improves on the previous V4.1-Flash across HLE, DeepSWE, and other reported benchmarks. Cross-model performance and cost depend on the specific workload and evaluation setup.
2. Platform Availability and Ecosystem Rollout
DeepSeek says V4.1-Flash-1017 is available through its website, app, and API.
- Web and App Access: The updated model is available on DeepSeek's official website and app.
- API Access: DeepSeek also lists the updated model as available through its API.
DeepSeek describes this release as a minor update to V4.1-Flash, reporting faster task completion and shorter outputs. Its published data do not establish that DeepSeek leads all competing models or that its per-task costs are lower under comparable conditions.