DeepSeek V4.1 Flash: DeepSWE 77.0 tops Opus 5, but Opus still leads several tests
Summarizes DeepSeek's own numbers: 77.0 on DeepSWE v1.1 versus 74.2 before, 74.0 for Opus 5 and 73.0 for Sol, while Opus 5 still leads HLE and Terminal-Bench; MIT weights, unchanged API pricing, Search/Fetch at $0.75 per 1,000 calls.
DeepSeek releases V4.1-Flash-1017, an open model it says beats Claude Opus 5 on one big coding test (benchmarks are DeepSeek's own) DeepSeek-V4.1-Flash-1017 went live today on DeepSeek's website, app and API, announced at 3:35 PM BKK. It's an upgrade to V4.1-Flash. DeepSeek says it scores higher on benchmarks, finishes tasks faster and gives shorter answers On the DeepSWE v1.1 coding benchmark, DeepSeek's own table shows 77.0, compared with 74.2 for the previous version, 74.0 for Claude Opus 5 and 73.0 for GPT-5.6 Sol It doesn't win everything: by DeepSeek's own numbers, Opus 5 still leads on Humanity's Last Exam, Terminal-Bench and two other coding benchmarks The weights are open under the MIT licence on Hugging Face and ModelScope API pricing is unchanged at $0.30 per million input tokens and $1.20 per million output tokens at peak times, with half-price rates off-peak DeepSeek also opened its Search and Fetch APIs to all developers at $0.75 per 1,000 calls each ⚠️ The benchmark figures come from DeepSeek and haven't been independently tested yet. Would a cheap open model like this get you to switch from Claude or GPT for coding? 👇 #DeepSeek #AI #OpenSource #LLM #AIcoding