Deepseek has released Deepseek-V4.1-Flash-1017, an update to its Flash model that the company says handles coding and AI agent tasks better while using fewer tokens. The Chinese AI company has also opened its in-house web search tools to outside developers for the first time.
The model is available through Deepseek’s website, app, and API. Its open-source weights are on Hugging Face and ModelScope. Deepseek says it scores higher than V4.1-Flash, completes tasks faster, and gives shorter answers. The company describes the performance and cost tradeoff as a Pareto improvement.
Deepseek reports higher coding scores
To train the model for building complete software projects from scratch, Deepseek uses reinforcement learning with self-feedback, or RLSF. The method uses the model’s own confidence scores or critiques as a reward signal, without human evaluators or an external reward model. Deepseek says it helps with projects including games, 3D modeling, and design work. The model can also animate short explainer videos, similar to the recently released Opus 5.5.
Deepseek tuned the model for Deepseek Harness, Tencent CodeBuddy, and WorkBuddy. It recommends Deepseek Harness version v0.2.1-rc.1 or later.
In Deepseek’s benchmarks, V4.1-Flash-1017 beats V4.1-Flash and scores ahead of larger models, including Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3, in several coding and agent categories.
On the company’s DeepSWE cost chart, the updated model reaches 77.0 percent at an average of just under $0.20 per task. Its low setting exceeds 72 percent at about $0.10. V4.1-Flash scores 74.2 percent at around $0.50 per run. GPT-6 Astra, Gemini 3.8 Flash, and Claude Opus 5 cost between $2 and more than $10 per task on the same chart, without matching the updated model’s score.
Those results have not yet been independently benchmarked. Benchmark scores also do not settle how developers will judge code quality in use. For teams comparing models for repeated coding tasks, Deepseek’s reported cost per task is a relevant measure, but the chart is company data.
Read nextBrave adds encrypted GLM 5.3 Flash chats to LeoSearch and Fetch APIs become available
Deepseek’s Search and Fetch APIs are now open to all developers. The company says they use its own search engine, which handles hundreds of millions of queries a day worldwide. Setup instructions are in its API documentation. Each endpoint costs $0.75 per 1,000 calls.
A separate uncertainty concerns future open-weight releases from China. The Chinese government reportedly plans tighter publication rules or a possible ban, with a model’s place among the 10 most capable systems one deciding factor. For now, Deepseek API users can access Search and Fetch free through server-side tools in Deepseek Harness, Claude Code, and Codex until October 31, 2026, at 4:00 PM UTC.
