Baseten

DeepSeek-V4-Pro

Made by DeepSeek. Released 2026-04-24. Accessibility not recorded. Cheapest standard rate we hold: $3.48 per million output tokens, no context split.

What it costs

TierKindInOutCached inFrom
standardtext$1.74$3.48$0.14*registry-single

per million tokens, US dollars. * marks a rate from a single registry or two that disagree — the best available number rather than a confirmed one. Every other row is confirmed against a seller or a licensed index.

What it scores, by house

BenchmarkHouseScoreEvaluatedScaffold
chess puzzlesEpoch AI0.1±0.032026-08-06high
chess puzzlesEpoch AI0.1±0.032026-08-06none
chess puzzlesEpoch AI0.2±0.042026-06-16max
frontiermath tier 4 v2Epoch AI0.0±0.022026-06-17max
frontiermath tiers 1 3 v2Epoch AI0.5±0.032026-06-17max
gpqa diamondEpoch AI0.7±0.032026-08-06none
gpqa diamondEpoch AI0.9±0.022026-08-06high
gpqa diamondEpoch AI0.9±0.022026-06-16max
mystery game puzzlesEpoch AI0.2±0.042026-08-27none
otis mock aime 2024 2025near its ceilingEpoch AI1.0±0.032026-08-06high
otis mock aime 2024 2025near its ceilingEpoch AI0.5±0.082026-08-06none
otis mock aime 2024 2025near its ceilingEpoch AI1.0±0.022026-06-17max
simpleqa verifiedEpoch AI0.5±0.022026-08-27max
swe bench verifiedEpoch AI0.8±0.022026-06-18max
ale benchcollected by Epoch AI522none
ale benchcollected by Epoch AI1006high
cl bench lifecollected by Epoch AI0.1high
critptcollected by Epoch AI0.1high
critptcollected by Epoch AI0.1max
dtbenchnear its ceilingcollected by Epoch AI0.9max
forecastbenchcollected by Epoch AI55.3unknown
frontiercodecollected by Epoch AI0.2mini-swe-agent
lmcacollected by Epoch AI41.2max
proofbenchcollected by Epoch AI0.2max
scicodecollected by Epoch AI0.5high
scicodecollected by Epoch AI0.5max
surface evolver benchcollected by Epoch AI0.4high
vending bench 2collected by Epoch AI3285unknown
webdev arenacollected by Epoch AI1464high
webdev arenacollected by Epoch AI1445none
webdev arenacollected by Epoch AI1464unknown
weirdmlcollected by Epoch AI0.5high
weirdmlcollected by Epoch AI0.5max

These scores are not comparable between houses and are not added up. A score is dated by when it was evaluated, not when the model was released, and a score without its scaffold is not a measurement — which is why both are printed.

Every rate we hold · Why there is no single best model · Everything from Baseten