DeepSeek’s V4.1-Flash-1017 beats Claude Opus 5 in 4 of 9 benchmarks, costs less than $0.20 per task
DeepSeek released V4.1-Flash-1017 on October 17, an update to its September model, available via web, app, and API. The company claims it outperforms its predecessor in all metrics and surpasses Claude Opus 5 and top GPT models in some coding and agentic tasks. DeepSeek’s own benchmarks show 4 wins over Opus 5 in areas like DeepSWE and Agents’ Last Exam, but 5 losses in others, including…
Key points
- DeepSeek’s V4.1-Flash-1017 wins 4 of 9 benchmarks against Claude Opus 5, per DeepSeek’s own data
- Costs less than $0.20 per programming task on DeepSWE, vs. $2–$10+ for Claude Opus 5
- U.S. federal agencies banned DeepSeek over data transfer risks to China; privacy policy confirms data stored in China
The model’s cost efficiency stands out: DeepSeek says it costs less than $0.20 per programming task on DeepSWE, compared to $2–$10+ for models like Claude Opus 5. API pricing is $0.30 per million input tokens (peak) and $1.20 per million output tokens, half the rate during off-peak hours. However, the U.S. Department of Homeland Security ordered federal agencies to remove DeepSeek products on October 19 due to data transfer risks to China. DeepSeek’s privacy policy confirms data is stored in China, raising privacy concerns for users handling sensitive information. The model’s search API is free until October 31–November 1, depending on the source.
The story so far
3 episodes →- DeepSeek’s V4.1-Flash-1017 beats Claude Opus 5 in 4 of 9 benchmarks, costs less than $0.20 per taskthis story
[AI Research File No. 07] DeepSeek's new model 'surpasses Claude Opus 5': How many wins and losses are there across the 9 published metrics?
note.com · 24 October 2026
Loading the full article…
This text was published by note.com and written by 真少年. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- sFNX runs Qwen3.8-Flash-Next at 382 tokens/s using three used V100 GPUs · 1 src
- OpenAI announces GPT-6.1 Sol with claimed 86.6% to 97.9% COBOL migration accuracy · 8 src
- Moonshot AI releases Kimi K3.1 Pro model with 3 trillion parameters · 1 src
- CodeRabbit uses GPT‑6.1 Sol in security agent · 1 src
- Anthropic reports 3.6x jump in token volume after Claude Opus 5.5 launch · 3 src
Comments
via GitHub Discussions