DeepSeek 网摘Five Weeks with DeepSeek V4.1-Flash: Price, Capability, Stability
DeepSeek-V4.1-Flash shipped on September 10, 2026 with a 70% output-price cut and native vision input. About five weeks in (as of October 16), here is a stage review from public data — no benchmarks, just price, capability and stability.
Price: still the low anchor
Per the official table verified October 16, Flash runs $0.15/1M input (cache miss) and $0.60/1M output off-peak, with cache hits at $0.003; peak doubles those. Consistent with the launch claim of a 70% output cut, no walk-back in five weeks.
Capability: 1M context and native vision
The 1M-token context, 384K max output and native image input are its differentiators at this price: long-doc Q&A, codebase understanding and vision tasks (homework, receipts) in one model. Tool calls, JSON output and default-on thinking mode round out the list.
Ecosystem: free access and heat
On October 15 the open-source coding tool OpenCode made V4.1-Flash free to use; an October 8 Hacker News post titled "Why isn't the industry freaking out about DeepSeek 4.1 Flash" drew 1,100+ points, debating whether its value is already priced in. Earlier in October, reports put DeepSeek's API usage at three times OpenAI's.
Stability: one near-five-hour outage
On October 15, 00:11–05:04 UTC, web and API suffered partial unavailability for nearly five hours, with both Flash and Pro API services affected — see the outage post-mortem. The official status page reports July–October uptime of 99.46% for Flash API and 99.74% for Pro. Availability-sensitive systems should keep a cross-provider fallback.
Verdict
Five weeks in: price and capability held their launch promises, the ecosystem keeps widening, and stability is the watch item. See the launch review and the migration guide.