DeepSeek upgrades its low-cost AI model
Nikkei Asia reports an upgraded DeepSeek model that the company says beats OpenAI and Anthropic tools at lower cost, with no figures given.
In short
Nikkei Asia reported on October 19, 2026 that DeepSeek has shipped a revised version of the AI model it introduced in September, claiming it outperforms US tools from OpenAI and Anthropic at a lower cost.
At a glance
- Source: Nikkei Asia report dated October 19, 2026, written by Itsuro Fujino.
- What is new: an upgraded version of the AI model DeepSeek introduced in September.
- Company claim: stronger performance than US models from OpenAI and Anthropic at lower cost.
- Not in the readable part of the report: model name, prices, benchmark scores, parameter count.
Nikkei Asia reported on October 19, 2026 that DeepSeek has pushed out a revised version of the AI model it introduced in September. The company says the update outperforms US models from OpenAI and Anthropic while costing less. Nothing in the publicly readable part of the report supports that with figures.
What is actually confirmed
Three things hold up. The report is dated October 19, 2026 and carries Itsuro Fujino's byline. It describes an improvement to an existing September model rather than a new family. And it frames the pitch the way DeepSeek has consistently framed its work: cheaper, without conceding performance.
What is missing
There is no model name in the readable text, no token pricing, no benchmark scores, no parameter count and no context window. Without those, the performance comparison is a vendor statement rather than a measurement. Testing the cost claim needs a published price sheet and a named evaluation, and we have neither.
The second half of the headline — doubts circulating in the tech sector — is also unresolved in what we could read. We cannot say which doubts, whose they are, or what they concern. So we are flagging that as a gap instead of filling it with background.
Why the price line matters
Cost is DeepSeek's central argument, which makes it the fair yardstick for this release. The useful check is not the announcement but whether pricing and reproducible evaluations follow. Until they do, this is a claim rather than evidence.
What we check next
We will update this article once a verifiable source provides the model name, the prices and the benchmark numbers. For now it stands as a company claim, reported by Nikkei Asia, with the numbers still absent.
FAQ
What is new in DeepSeek's latest AI model?
Nikkei Asia describes it as an improved version of a model DeepSeek introduced in September. The readable part of the report gives no model name.
Is DeepSeek cheaper than OpenAI and Anthropic?
That is the company's own claim so far. Without published per-token prices and a named head-to-head evaluation, the claim cannot be checked.
What numbers are known about the new DeepSeek model?
None. Prices, benchmark scores, parameter count and context window are absent from the readable part of the report, and we will add them only from a verifiable source.
