DeepSeek's free model update chases Claude on coding

Corrected 19 Oct: The sources never name Claude; 'chases Claude' was our inference from the Hugging Face page, which lists Opus 5 at 74.0 vs DeepSeek's 77.0 on DeepSWE (DeepSeek's own table). All corrections

DeepSeek has released an update to its model called V4.1-Flash-1017. The weights are open, and by DeepSeek's own table it scores 77.0 on a coding test, ahead of a model listed as Opus 5 on that one test. That matters because a free model is now close to the paid leaders on coding.

What happened

On 17 October 2026, DeepSeek released DeepSeek-V4.1-Flash-1017. Its release note says the model is live on its website, its app and its API. The note says it has "stronger benchmark performance, faster task completion, and shorter outputs" than the version before it.

The weights are posted on Hugging Face and ModelScope, two sites where people share AI models. The Hugging Face page calls the release a "minor update" and lists an MIT license for the weights. That is a very open license.

DeepSeek also says it trained the model with a method where the model learns from its own feedback. The goal was to help it build whole projects, such as games and 3D scenes, per the release note.

Key facts

• Coding score: 77.0 on DeepSWE v1.1, a test of fixing real software problems, according to DeepSeek's Hugging Face page. The previous version scored 74.2.
• Rivals on that test, from the same table: 74.0 for Opus 5 and 73.0 for GPT-5.6 Sol. These are DeepSeek's own numbers, not an outside test.
• Where it trails: on most other rows, the table shows Opus 5 ahead, including Terminal-Bench 4.0 (51.8 against 42.3) and the HLE reasoning test (56.3 against 42.6).
• Size: 763 billion parameters in total, the Hugging Face page says. Parameters are the numbers inside a model that hold what it learned.
• Memory: Hugging Face lists 1 million tokens as the recommended context window setting, which means it can read very long inputs.
• Where to use it: the release note says it is tuned for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy.

Background

This update is a small step on top of an earlier model, V4.1-Flash. The Hugging Face page lists that older model's scores next to the new ones, which makes the gain easy to see. On DeepSWE it went from 74.2 to 77.0.

Techmeme collected early coverage and comments. The posts and stories it links are mostly social media posts and secondary reports, not independent tests. One early social media post it links to reported a lower score for the model on a hidden-bug test than for three rivals.

What it means for you

If you code with AI, you now have a free option that DeepSeek says is near the top. You can try it through DeepSeek's own tools or run it yourself if you have strong hardware.

Our take: a score on one test is not the same as a better tool. The model leads on DeepSWE but trails Opus 5 on most other rows in the same table. Try it on your own work before you switch.

What happens next

The release note says billing for server-side tool calls starts October 31, 2026. The Hugging Face page points to a technical report for readers who want to see how the model was built.

Our take: watch for tests from outside groups. They will show if the 77.0 holds up.

FAQ

Is DeepSeek V4.1-Flash-1017 free?

The weights are open under an MIT license, per the Hugging Face page. Using DeepSeek's own API may cost money.

Is it better than Claude at coding?

Our take: not across the board. DeepSeek's table has it ahead on one coding test, DeepSWE, and behind Opus 5 on several others. The table does not say who makes Opus 5. Our take: it appears to be the Claude Opus line, but DeepSeek's page does not confirm that.

Who ran these tests?

DeepSeek published the table. The Techmeme roundup links mostly to social media posts and secondary reports, not independent tests.

Sources

• DeepSeek
• techmeme.com
• huggingface.co