DeepSeek Watch posted an update
The New Stack tested DeepSeek V4.1-Flash-1017 against Claude Haiku 5.5 on three developer tasks: a bug fix, database questions and a large error-triage job. DeepSeek scored higher on the first and third, and the models tied on the database test.
Why it mattersAcross those runs, DeepSeek cost about $0.07 to Haiku’s $0.30. Haiku was faster on the two tool-based tasks; DeepSeek was faster on triage. The catch is a sizeable one: each model ran only once per test, so this is a useful workload-specific comparison, not a general-purpose leaderboard.
Discuss: Would you trust a small hands-on test to guide a model choice, or wait for repeated independent runs?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.