DeepSeek V4.1 Flash 1017: faster coding agents at $0.20 per task
DeepSeek shipped a refreshed version of its fast model, DeepSeek V4.1 Flash 1017, claiming faster task completion, shorter outputs, and a steep cut in cost per coding job. The weights are free to download under the MIT License from Hugging Face and ModelScope.
What Changed
The update targets reasoning, coding, agent tasks, and visual benchmarks. Shorter outputs are the key cost lever: you pay per token, so a model that finishes the same job with fewer tokens runs cheaper at scale.
For full software project generation, DeepSeek used a training method called reinforcement learning with self-feedback. The model uses its own confidence scores or self-written critiques as the reward signal instead of human graders. DeepSeek says this helped on tasks from simple games to 3D modeling and design work.
The model is tuned for DeepSeek’s own Harness tool (version v0.2.1-rc.1 or later), plus Tencent’s CodeBuddy and WorkBuddy apps.
The Benchmark Numbers
On the DeepSWE coding benchmark, DeepSeek reports the new model scores 77.0 percent at an average of just under $0.20 per task. The previous version scored 74.2 percent at roughly $0.50 per task. DeepSeek also claims the model beats Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3 on several coding and agent categories.
These are the company’s own numbers. Independent results have not been published yet. Benchmark scores also don’t always reflect how clean or maintainable the code is in real projects.
New Search and Fetch APIs
Alongside the model, DeepSeek opened its in-house search engine to developers for the first time via two new tools: a Search API that returns structured web results and a Fetch API that retrieves a single page as Markdown. The company says the search engine already handles hundreds of millions of queries per day.
- Pricing: $0.75 per 1,000 calls for each endpoint
- Failed requests, empty results, and timeouts are not billed
- Rate limits are set per account, not per API key
- Server-side tool calls via DeepSeek Harness, Claude Code, or Codex are free until billing starts at midnight Beijing time on November 1, 2026
⚠️ One Risk to Watch
The weights ship without a standard chat template file. Instead, DeepSeek includes a Python reference encoder and a new toolkit called deepseek-recipe, a set of Rust libraries with Python bindings that converts common API request formats into DeepSeek’s prompt format.
China is also reportedly weighing tighter rules on publishing powerful open-weight models. If those rules land, future releases from Chinese labs could look very different.
What to Do With This
- Run your own tests on real tasks before trusting the benchmark claims
- Compare cost per finished task, not just price per token
- Try the Search and Fetch APIs before November 1 while some tool calls are free
- Check your company and legal policies before sending code or data to a China-based API
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
