Skip to content
MaungaAI news, explained

Models

DeepSeek says its cheap model beats Claude in four tests and loses in five

DeepSeek has improved its budget model and, according to its own figures, it now competes with Claude Opus 5 at programming. It is also opening a service to everyone so their programs can search the web.

1 min read3 sourcesWritten with AI · machine-translated

Share: WhatsApp · Email

AI-generated illustration for Maunga. Not a photograph.

The essentials

  • DeepSeek has released DeepSeek-V4.1-Flash-1017, an improved version of the V4.1-Flash model it launched in September (MarkTechPost, TLDR AI).
  • Claimed by one partyAccording to DeepSeek's figures, it beats Claude Opus 5 in four of nine tests, such as one on programming, and loses in the other five (MarkTechPost).
  • Claimed by one partyAccording to MarkTechPost, DeepSeek charges $0.75 for every 1,000 successful requests to its web search, and the same for its page reader (MarkTechPost).

Why it matters

If the figures hold up, a model built to be cheap is getting close to Claude Opus 5 at programming. Also, people who build agents (AI programs that do tasks on their own) can search the internet with the same DeepSeek account, without signing up with another provider.

What does it mean for me?

If you use the DeepSeek app or website, you already have this version, according to MarkTechPost.

The details

According to MarkTechPost, the model already works on DeepSeek's website, app and API. At peak hours it costs $0.30 per million input tokens (chunks of text) that are not stored in cache, and $1.20 per million output tokens. Outside those hours it costs half. Using search from DeepSeek Harness, Claude Code or Codex is free until November 1, 2026 (MarkTechPost).

What we don't know

  • Whether anyone independent has checked the results: DeepSeek publishes all of them.
  • Whether DeepSeek will release this version so anyone can download it and use it freely. MarkTechPost says it has not made this clear, although TLDR AI links to a page for the model on Hugging Face.

Sources

This story is based on 3 sources. In the text, each fact names in brackets who reports it.

Was this story useful? Yes Not really

Spanish original

The thread: DeepSeek · DeepSeek-V4.1-Flash-1017

  1. DeepSeek's agent that rarely interrupts has only been tested in simulation
  2. US accuses six Chinese companies of copying its AI and proposes throwing them off
  3. DeepSeek releases its AI engine, and Latent Space estimates an 80% margin
  4. DeepSeek says its cheap model beats Claude in four tests and loses in five · you are here