Key points
- Deepseek released V4.1-Flash-1017 as an open-weight model on Hugging Face and ModelScope, also available in its app, website and API.
- On DeepSWE, Deepseek's own chart shows 77.0 percent at just under $0.20 per task, versus 74.2 percent at about $0.50 for its predecessor.
- Search and Fetch APIs, built on Deepseek's internal search engine, are now open to all developers at $0.75 per 1,000 calls per endpoint.
- Independent benchmarks are not yet available; all performance figures come from Deepseek.
Why it matters for marketers: Cheaper per-task agent runs and a new paid search API could lower costs for marketing automation, though the numbers are vendor-reported and unverified.
Deepseek has released V4.1-Flash-1017, an update to its Flash model, through its website, app and API. The weights are also published openly on Hugging Face and ModelScope. The company says the model is faster than its predecessor, writes shorter outputs and handles end-to-end software projects and agent tasks better, thanks to a post-training method that uses the model's own feedback as a reward signal.
The cost figures are the headline for anyone running agents at scale. In Deepseek's own DeepSWE chart, the model reaches 77.0 percent for just under $0.20 per task, while a lower setting still tops 72 percent for roughly $0.10. The prior version scored 74.2 percent at around $0.50. Deepseek's chart places rivals including Claude Opus 5, GPT-6 Astra and Gemini 3.8 Flash at $2 to over $10 per task, with lower accuracy.
Deepseek is also exposing its in-house search engine, which it says serves hundreds of millions of queries daily, through Search and Fetch APIs. Each endpoint costs $0.75 per 1,000 calls. Until October 31, 2026, API users can try these features at no charge inside Deepseek Harness, Claude Code and Codex.
For marketing and content teams building research, monitoring or content-production agents, lower per-task cost and a bundled search tool could make multi-step workflows cheaper to run. The caveat is that all results are Deepseek's own, and independent testing has not appeared. Benchmark scores also often differ from how output quality feels in real use.
Watch for outside evaluations, and for China's reported plans to tighten rules on releasing open-weight models, which could affect how safely teams can depend on Chinese models in production.
Sources
This brief is our own summary. Read the full reporting at the source.
