Epoch

gpt-5.6-terra

Made by OpenAI. Released 2026-07-09. Accessibility not recorded. Cheapest standard rate we hold: $12 per million output tokens, at short context.

What it costs

TierKindInOutCached inFrom
batchtext · long$2$9$0.20seller
batchtext · short$1$6$0.10seller
fasttext · long$8$36$0.80seller
fasttext · short$4$24$0.40seller
flextext · long$2$9$0.20seller
flextext · short$1$6$0.10seller
standardtext · long$4$18$0.40seller
standardtext · short$2$12$0.20seller

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±0.022026-08-07none
chess puzzlesEpoch AI0.2±0.042026-08-07low
chess puzzlesEpoch AI0.5±0.052026-07-09max
frontiermath tier 4 v2Epoch AI0.7±0.072026-07-09max
frontiermath tiers 1 3 v2Epoch AI0.9±0.022026-07-09max
gpqa diamondEpoch AI0.9±0.022026-08-07low
gpqa diamondEpoch AI0.8±0.032026-08-07none
gpqa diamondEpoch AI0.9±0.022026-07-09max
mystery game puzzlesEpoch AI0.1±0.032026-08-28medium
mystery game puzzlesEpoch AI0.1±0.032026-08-27none
mystery game puzzlesEpoch AI0.2±0.042026-08-27low
mystery game puzzlesEpoch AI0.3±0.052026-07-28max
otis mock aime 2024 2025near its ceilingEpoch AI0.5±0.082026-08-07none
otis mock aime 2024 2025near its ceilingEpoch AI0.9±0.052026-08-07low
otis mock aime 2024 2025near its ceilingEpoch AI1.0±0.002026-07-09max
simpleqa verifiedEpoch AI0.4±0.022026-08-10max
ale benchcollected by Epoch AI1951max
arc aginear its ceilingcollected by Epoch AI0.9xhigh
arc aginear its ceilingcollected by Epoch AI0.9high
arc aginear its ceilingcollected by Epoch AI0.6low
arc aginear its ceilingcollected by Epoch AI0.8medium
arc aginear its ceilingcollected by Epoch AI1.0max
arc agi 2collected by Epoch AI0.7xhigh
arc agi 2collected by Epoch AI0.2low
arc agi 2collected by Epoch AI0.8max
arc agi 2collected by Epoch AI0.4medium
arc agi 2collected by Epoch AI0.7high
blueprint bench 2collected by Epoch AI0.3unknown
critptcollected by Epoch AI0.1low
critptcollected by Epoch AI0.2medium
critptcollected by Epoch AI0.2high
critptcollected by Epoch AI0.3max
critptcollected by Epoch AI0.0none
critptcollected by Epoch AI0.3xhigh
cursorbenchcollected by Epoch AI0.5low
cursorbenchcollected by Epoch AI0.6xhigh
cursorbenchcollected by Epoch AI0.5high
cursorbenchcollected by Epoch AI0.5medium
cursorbenchcollected by Epoch AI0.6max
deepswecollected by Epoch AI0.7mini-swe-agent
dtbenchnear its ceilingcollected by Epoch AI0.9max
frontiercodecollected by Epoch AI0.4codex
gdp pdfcollected by Epoch AI0.2unknown
gdp pdfcollected by Epoch AI0.2medium
lmcacollected by Epoch AI52.2max
proofbenchcollected by Epoch AI0.7xhigh
scicodecollected by Epoch AI0.5low
scicodecollected by Epoch AI0.5medium
scicodecollected by Epoch AI0.4none
scicodecollected by Epoch AI0.5xhigh
scicodecollected by Epoch AI0.5high
scicodecollected by Epoch AI0.5max
simplebenchcollected by Epoch AI0.5xhigh
simplebenchcollected by Epoch AI0.5unknown
surface evolver benchcollected by Epoch AI0.8xhigh
vending bench 2collected by Epoch AI7343unknown
webdev arenacollected by Epoch AI1520xhigh
weirdmlcollected by Epoch AI0.8high

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 Epoch