A Chinese AI lab has just upgraded its budget model, and on one coding test it now scores…
A Chinese AI lab has just upgraded its budget model, and on one coding test it now scores above Anthropic's Claude Opus 5.
DeepSeek released V4.1-Flash-1017 on 17 October. Its API price hasn't changed, and the web app and phone app stay free. On the DeepSWE coding benchmark it reports 77.0; a third-party comparison puts Claude Opus 5 at 74.0.
Some honest caveats:
・These are DeepSeek's own numbers, not independently checked.
・On the other three tests in the same table (Terminal-Bench, NL2Repo, HLE), Opus 5 still leads.
・DeepSeek's API stores data on servers in mainland China, so think twice before sending it client data.
I use AI agents every day, and for me the question is no longer "which AI is smartest". As models keep getting cheaper and the gap keeps closing, what are we actually paying for? Speed, reliability, or knowing where our data lives and who answers when something goes wrong?
What matters most to you when you choose an AI tool? Tell me what you think in the comments 👇
Sources:
https://api-docs.deepseek.com/updates/
https://kingy.ai/blog/deepseek-v4-1-flash-1017-specs-benchmarks-pricing/
