DeepSeek’s Flash update lifts coding scores, but its own tests still favour Claude in most benchmarks

DeepSeek’s V4.1-Flash-1017 update improves on its predecessor in several coding and agent benchmarks, while keeping the same API pricing and model identifier. The practical catch: DeepSeek’s own tests still put Claude Opus 5 ahead in seven of the 11 comparisons.

DeepSeek Watch analysis

What happened

DeepSeek announced the update on 17 October, and it is now the default Flash model in the company’s web and mobile products, according to Incrypted’s overview and hands-on guide. The report says API users do not need to change their code: the existing deepseek-flash identifier serves the updated model.

In DeepSeek’s internal benchmarks, DeepSWE v1.1 rose from 74.2 to 77.0, while Terminal-Bench 4.0 went from 31.2 to 42.3. Incrypted says the new version beat Claude Opus 5 in four of the 11 tests, but trailed it in the other seven. Those are company-run results, not independent measurements.

Why it matters

The update appears most relevant to developers using agents for coding and other multi-step tasks. They get the new version through the existing API route, with pricing unchanged, rather than having to rebuild around a new model name.

The benchmark gains are not a blanket verdict that Flash now leads. The report describes stronger results in some agentic and coding tests, alongside weaker performance in others. Useful signal, then, but not a universal leaderboard victory lap.

Our read

The strongest takeaway is the combination of a meaningful Terminal-Bench jump and a no-migration API update. That makes the model worth another look for existing users, particularly those running coding agents. Treat the performance claims as a reason to test it on your own workload, not as a substitute for doing so. Benchmarks are helpful; they are not your bug tracker.

What to watch

• Whether independent tests reproduce the gains on coding and agent tasks.
• How the updated model performs on real workloads at unchanged API pricing.
• Whether DeepSeek follows through on its previously stated plan for V4.1-Pro.

Discussion spark: Would a sizeable gain on one coding benchmark persuade you to switch models, or should independent tests across your own workload be the deciding factor?

not affiliated with or endorsed by DeepSeek