\n DeepSeek V4.1 Flash\n

\n Observed performance across providers serving{\" \"}\n \n deepseek/deepseek-v4.1-flash\n {\" \"}\n through Vercel AI Gateway. This benchmark sends the same long\n multi-turn prefix four times to each provider.\n

\n \n View on Vercel ↗\n \n \n View benchmark code\n \n

\n {error && (\n

\n {error}\n

\n )}\n\n {!results\n ?

Loading results…

\n : (\n <>\n

\n Routing strategies\n

\n Ways of calling the gateway instead of pinning one\n provider. Every strategy is restricted with{\" \"}\n only to the\n US-based providers below and sends{\" \"}\n zeroDataRetention: true{\" \"}\n and{\" \"}\n disallowPromptTraining: true. They\n differ only in how the gateway orders those providers. The\n line under each name shows the routing hints, which\n provider actually served the request, and how often the\n gateway failed over.\n

\n \n p.kind === \"strategy\"\n )}\n emptyMessage=\"No strategy runs yet — the next scheduled run will add them.\"\n />\n

\n\n

\n Providers\n

\n {results.runCount} completed{\" \"}\n {results.runCount === 1 ? \"run\" : \"runs\"} ·{\" \"}\n {results.attemptCount} measurements · last run{\" \"}\n {formatTimestamp(results.lastCompletedAt)} ·{\" \"}\n \n {results.benchmarkVersion}\n \n

\n \n p.kind !== \"strategy\"\n )}\n />\n

\n \n )}\n