Anthropic

Claude Opus 5.5

Made by Anthropic. Released 2026-09-22. Accessibility not recorded. Cheapest standard rate we hold: $20 per million output tokens, no context split.

What it costs

TierKindInOutCached inFrom
batchtext$2$10—seller
fasttext$8$40—seller
standardtext$4$20$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 AI0.7±0.052026-10-13max
ebr benchEpoch AI0.7±0.072026-09-22max
frontiermath erdosEpoch AI0.0±0.022026-09-24max
frontiermath tier 4 v2Epoch AI0.9±0.042026-09-22max
frontiermath tiers 1 3 v2Epoch AI0.9±0.022026-09-22max
furniture assemblyEpoch AI0.8±0.052026-09-22max
gpqa diamondEpoch AI0.9±0.022026-09-22max
mirrorcodeEpoch AI1.0±0.032026-10-30max
mirrorcodeEpoch AI0.8±0.082026-09-22high
mystery game puzzlesEpoch AI0.7±0.052026-09-29max
otis mock aime 2024 2025near its ceilingEpoch AI1.0±0.002026-09-22max
simpleqa verifiedEpoch AI0.7±0.012026-09-22max
spireclimbEpoch AI0.1±0.022026-11-04max
spireclimb sts1 a20 heartEpoch AI0.1±0.032026-11-04max
spireclimb sts2 a10Epoch AI0.1±0.032026-09-23max
ale benchcollected by Epoch AI2147—high
apex agentscollected by Epoch AI0.7—medium
apex agentscollected by Epoch AI0.8—max
arc aginear its ceilingcollected by Epoch AI1.0—xhigh
arc aginear its ceilingcollected by Epoch AI0.9—low
arc aginear its ceilingcollected by Epoch AI1.0—max
arc aginear its ceilingcollected by Epoch AI1.0—high
arc aginear its ceilingcollected by Epoch AI1.0—medium
arc agi 2collected by Epoch AI0.9—max
arc agi 2collected by Epoch AI0.9—medium
arc agi 2collected by Epoch AI0.7—low
arc agi 2collected by Epoch AI0.9—xhigh
arc agi 2collected by Epoch AI0.9—high
blueprint bench 2collected by Epoch AI0.5—unknown
critptcollected by Epoch AI0.3—xhigh
critptcollected by Epoch AI0.2—low
critptcollected by Epoch AI0.3—high
critptcollected by Epoch AI0.3—medium
critptcollected by Epoch AI0.3—max
cursorbenchcollected by Epoch AI0.5—medium
cursorbenchcollected by Epoch AI0.4—low
cursorbenchcollected by Epoch AI0.6—max
cursorbenchcollected by Epoch AI0.6—xhigh
cursorbenchcollected by Epoch AI0.6—high
dtbenchnear its ceilingcollected by Epoch AI1.0—low
dtbenchnear its ceilingcollected by Epoch AI1.0—max
dtbenchnear its ceilingcollected by Epoch AI1.0—xhigh
dtbenchnear its ceilingcollected by Epoch AI1.0—high
dtbenchnear its ceilingcollected by Epoch AI1.0—medium
frontiercodecollected by Epoch AI0.5—claude-code
frontierswecollected by Epoch AI0.6—proximus
gdp pdfcollected by Epoch AI0.3—max
lmcacollected by Epoch AI65.7—medium
lmcacollected by Epoch AI65.9—high
lmcacollected by Epoch AI64.9—low
lmcacollected by Epoch AI68.2—max
lmcacollected by Epoch AI66.2—xhigh
posttrainbenchcollected by Epoch AI0.4—Claude Code
proofbenchcollected by Epoch AI1.0—max
rlicollected by Epoch AI0.2—unknown
scicodecollected by Epoch AI0.6—low
scicodecollected by Epoch AI0.7—max
scicodecollected by Epoch AI0.7—xhigh
scicodecollected by Epoch AI0.6—medium
scicodecollected by Epoch AI0.6—high
simplebenchcollected by Epoch AI0.9—unknown
vending bench 2collected by Epoch AI9235—unknown
webdev arenacollected by Epoch AI1820—max
weirdml v3collected by Epoch AI0.3—claude_code

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.

What retires, and when

  • deprecation — 2027-09-22 (Active)

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