PRICE~deepseek/deepseek-v4-flash-latest output_per_mtok decreased from 1.04 to 0.6018mPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok increased from 0.0038 to 0.01318mPRICE~deepseek/deepseek-flash-latest cached_input_per_mtok decreased from 0.0021 to 0.002118mPRICEdeepseek/deepseek-v4.1-flash cached_input_per_mtok increased from 0.0021 to 0.00618mPRICEdeepseek/deepseek-v4.1-flash output_per_mtok increased from 0.68 to 1.2018mPRICEdeepseek/deepseek-v4.1-flash input_per_mtok increased from 0.015 to 0.3018mPRICE~deepseek/deepseek-v4-flash-latest cached_input_per_mtok decreased from 0.0077 to 0.001149mPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok decreased from 1.28 to 1.0449mPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.0077 to 0.003849mPRICE~deepseek/deepseek-flash-latest cached_input_per_mtok decreased from 0.02 to 0.002149mPRICE~deepseek/deepseek-flash-latest output_per_mtok increased from 0.60 to 0.6849mPRICE~deepseek/deepseek-flash-latest input_per_mtok decreased from 0.02 to 0.01549mPRICEdeepseek/deepseek-v4.1-flash cached_input_per_mtok decreased from 0.02 to 0.002149mPRICEdeepseek/deepseek-v4.1-flash output_per_mtok increased from 0.60 to 0.6849mPRICEdeepseek/deepseek-v4.1-flash input_per_mtok decreased from 0.02 to 0.01549mPRICEdeepseek/deepseek-v4-flash-0731 cached_input_per_mtok increased from 0.0077 to 0.01749mPRICEdeepseek/deepseek-v4-flash-0731 input_per_mtok increased from 0.0077 to 0.01749mPRICE~deepseek/deepseek-v4-flash-latest cached_input_per_mtok decreased from 0.012 to 0.00771hPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.012 to 0.00771hPRICEdeepseek/deepseek-v4-flash-0731 cached_input_per_mtok decreased from 0.012 to 0.00771hPRICEdeepseek/deepseek-v4-flash-0731 input_per_mtok decreased from 0.012 to 0.00771hPRICEdeepseek/deepseek-v3.1-terminus input_per_mtok decreased from 0.30 to 0.271hPRICE~deepseek/deepseek-v4-flash-latest cached_input_per_mtok increased from 0.0011 to 0.0122hPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok increased from 0.60 to 1.282hPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.013 to 0.0122hPRICEdeepseek/deepseek-v4-flash-0731 cached_input_per_mtok decreased from 0.017 to 0.0122hPRICEdeepseek/deepseek-v4-flash-0731 input_per_mtok decreased from 0.017 to 0.0122hNEW MODELnvidia/switchyard appeared in the catalogue2hPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok decreased from 1.04 to 0.602hPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok increased from 0.0038 to 0.0132hPRICEnvidia/nemotron-3-ultra-550b-a55b cached_input_per_mtok decreased from 0.12 to 0.102hPRICEnvidia/nemotron-3-ultra-550b-a55b output_per_mtok decreased from 2.40 to 2.202hPRICEnvidia/nemotron-3-ultra-550b-a55b input_per_mtok decreased from 0.60 to 0.502hPRICEdeepseek/deepseek-v4-flash-0731 cached_input_per_mtok increased from 0.0051 to 0.0172hPRICEdeepseek/deepseek-v4-flash-0731 input_per_mtok increased from 0.0051 to 0.0172hPRICE~moonshotai/kimi-latest cached_input_per_mtok decreased from 0.80 to 0.293hPRICE~moonshotai/kimi-latest output_per_mtok decreased from 13.00 to 11.363hPRICE~moonshotai/kimi-latest input_per_mtok decreased from 1.39 to 1.003hPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok increased from 0.95 to 1.043hPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.0058 to 0.00383h

The data is yours

Everything the benchmark publishes is available raw: JSON or CSV, no key, no signup, CC BY 4.0 with a link back. Numbers come from the same engine and the same 60-second cache the site itself renders from.

Endpoints

EndpointReturnsCSV
/api/v0/leaderboardThe aggregated provider ranking, exactly as the homepage computes it.?format=csv
/api/v0/pairsEvery measured provider × model pair with its current canonical metrics — the raw table behind everything.?format=csv
/api/v0/eventsThe market event stream: detected performance shifts, incidents and price changes.—

Try it: curl -s https://inferencebench.dev/api/v0/leaderboard | jq .leaderboard[0]

Terms, honestly short

License: CC BY 4.0 — use it, republish it, build on it; cite InferenceBench with a link. Every figure is an independent measurement from the canonical European vantage (Falkenstein, DE); read the methodology before drawing conclusions the data does not support. Fields can be null where nothing is measured — absence is information here, never invention.