Anthropic

claude-sonnet-4-20250514

Made by Anthropic. Released 2025-05-22. Accessibility not recorded. Cheapest standard rate we hold: $15 per million output tokens, no context split.

What it costs

TierKindInOutCached inFrom
standardtext$3$15$0.30licensed-index

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
frontiermathEpoch AI0.0±0.012025-07-04
frontiermath tier 4Epoch AI0.0±0.002025-07-01
gpqa diamondEpoch AI0.8±0.032025-05-2659K
gpqa diamondEpoch AI0.8±0.032025-05-2232K
gpqa diamondEpoch AI0.7±0.032025-05-22
gpqa diamondEpoch AI0.8±0.032025-05-2216K
math level 5near its ceilingEpoch AI0.8±0.012025-05-22
otis mock aime 2024 2025near its ceilingEpoch AI0.7±0.072025-05-2359K
otis mock aime 2024 2025near its ceilingEpoch AI0.5±0.082025-05-2216K
otis mock aime 2024 2025near its ceilingEpoch AI0.3±0.072025-05-22
otis mock aime 2024 2025near its ceilingEpoch AI0.7±0.072025-05-2232K
aider polyglotcollected by Epoch AI56.4
aider polyglotcollected by Epoch AI61.332K
ale benchcollected by Epoch AI65532K
apex agentscollected by Epoch AI0.2unknown
arc aginear its ceilingcollected by Epoch AI0.416K
arc aginear its ceilingcollected by Epoch AI0.2
arc aginear its ceilingcollected by Epoch AI0.31K
arc aginear its ceilingcollected by Epoch AI0.38K
arc agi 2collected by Epoch AI0.08K
arc agi 2collected by Epoch AI0.0
arc agi 2collected by Epoch AI0.116K
arc agi 2collected by Epoch AI0.01K
critptcollected by Epoch AI0.0unknown
cybenchcollected by Epoch AI0.3
deepresearchbenchcollected by Epoch AI0.52K
dtbenchnear its ceilingcollected by Epoch AI0.8unknown
enigma evalcollected by Epoch AI0.0unknown
fictionlivebenchcollected by Epoch AI0.4
forecastbenchcollected by Epoch AI60.2unknown
geobenchcollected by Epoch AI2468
geobenchcollected by Epoch AI236932K
gsocollected by Epoch AI0.0OpenHands
hlecollected by Epoch AI0.1unknown
lech mazur writingcollected by Epoch AI8.116K
lech mazur writingcollected by Epoch AI8.1
lmcacollected by Epoch AI29.0unknown
metr time horizonscollected by Epoch AI74.916K
mindcubecollected by Epoch AI0.4unknown
os worldcollected by Epoch AI41.4claude-4-sonnet-20250514 (100 steps)
os worldcollected by Epoch AI43.9claude-4-sonnet-20250514 (50 steps)
os worldcollected by Epoch AI31.2claude-4-sonnet-20250514 (15 steps)
scicodecollected by Epoch AI0.4unknown
simplebenchcollected by Epoch AI0.5unknown
simplebenchcollected by Epoch AI0.512K
the agent companycollected by Epoch AI0.4
vpctcollected by Epoch AI0.3
vpctcollected by Epoch AI0.332K
weirdmlcollected by Epoch AI0.516K
weirdmlcollected by Epoch AI0.4

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

  • deprecation2026-06-15 (Retired)

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