Deepseek releases open-weight V4.1-Flash-1017 with gains in agentic coding and launches paid Search and Fetch APIs

Deepseek released Deepseek-V4.1-Flash-1017, an updated checkpoint of its Flash model. It is available on its website, app and API, with open weights on Hugging Face and ModelScope. Deepseek reports 77.0% on the DeepSWE coding benchmark at just under $0.20 per task, against 74.2% at about $0.50 for the previous version, with faster responses and shorter outputs. It also opened its in-house search engine to outside developers through paid Search and Fetch APIs at $0.75 per 1,000 calls each. Access is free through server-side tools in Deepseek Harness, Claude Code and Codex until 4:00 PM UTC on 31 October 2026.

Developers running agentic coding workloads get an open-weight model that, on Deepseek's own numbers, is cheaper per task and slightly better than its predecessor. That is worth a test if per-task cost matters to you. The Search and Fetch APIs give teams a new, priced option for web grounding, and the free window until 31 October lets them trial it at no cost. The change is incremental, though: it is a dated update to an existing model, not a new generation, and any decision to switch should wait for your own testing on your own tasks.

The claim that the smaller Flash model beats Claude Opus 5, GPT 5.6-Sol, Kimi K3 and GLM 5.3 rests on Deepseek's own benchmark runs, and the coverage says no independent results exist yet. Phrases like "Pareto improvement" and "once again massively cutting costs" are framing, not evidence, and a single benchmark with a 2.8 point gain is not proof of broad superiority. The article also does not say how comparison models were configured, or whether the cost figures include the retries and tool calls that real agent work needs.

1. 01Independent DeepSWE and other agentic coding results (for example from third-party leaderboards or evaluators) over the next two to four weeks, to see whether the 77.0% score and the lead over rivals hold up.
2. 02Real-world cost per completed task reported by developers using the open weights or the API, including retries and tool calls, compared with the quoted $0.20.
3. 03What Deepseek does with Search and Fetch pricing and availability after the free window ends on 31 October 2026, and whether usage or integrations beyond Deepseek Harness, Claude Code and Codex appear.