Google

Gemini 2.5 Pro

Made by Google DeepMind. Released 2025-06-17. Accessibility not recorded. Cheapest standard rate we hold: $10 per million output tokens, no context split.

What it costs

TierKindInOutCached inFrom
batchtext$0.63$5seller
flextext$0.63$5seller
prioritytext$2.25$18seller
standardtext$1.25$10seller

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 AI0.2±0.042025-12-08
frontiermathEpoch AI0.1±0.022025-11-24
frontiermath tier 4Epoch AI0.0±0.032025-07-03
frontiermath tier 4 v2Epoch AI0.0±0.002026-06-11
frontiermath tiers 1 3 v2Epoch AI0.2±0.032026-06-11
gpqa diamondEpoch AI0.9±0.022025-11-16
otis mock aime 2024 2025near its ceilingEpoch AI0.8±0.052025-11-16
swe bench verifiedEpoch AI0.6±0.022026-02-13
ale benchcollected by Epoch AI78632K
algotunecollected by Epoch AI1.5
apex agentscollected by Epoch AI0.2
arc aginear its ceilingcollected by Epoch AI0.38K
arc aginear its ceilingcollected by Epoch AI0.416K
arc aginear its ceilingcollected by Epoch AI0.432K
arc agi 2collected by Epoch AI0.016K
arc agi 2collected by Epoch AI0.032K
arc agi 2collected by Epoch AI0.08K
critptcollected by Epoch AI0.0
deepresearchbenchcollected by Epoch AI0.4
dtbenchnear its ceilingcollected by Epoch AI0.8
forecastbenchcollected by Epoch AI60.4
gdpvalcollected by Epoch AI0.2
gsocollected by Epoch AI0.0OpenHands
lech mazur writingcollected by Epoch AI8.4
lmcacollected by Epoch AI34.8
scicodecollected by Epoch AI0.4
spatialviz benchcollected by Epoch AI0.4
terminalbenchcollected by Epoch AI0.3Mini-SWE-Agent
terminalbenchcollected by Epoch AI0.3Terminus 2
terminalbenchcollected by Epoch AI0.2Gemini CLI
terminalbenchcollected by Epoch AI0.2OpenHands
vending bench 2collected by Epoch AI574
webdev arenacollected by Epoch AI1226
weirdmlcollected by Epoch AI0.516K

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 Google