Moonshot

kimi-k3

Made by Moonshot. Released 2026-07-16. Accessibility not recorded. Cheapest standard rate we hold: $15 per million output tokens, no context split.

What it costs

Tier	Kind	In	Out	Cached in	From
standard	text	$3	$15	$0.30	registry

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

Benchmark	House	Score	Evaluated	Scaffold
chess puzzles	Epoch AI	0.4±0.05	2026-10-13	high
chess puzzles	Epoch AI	0.4±0.05	2026-10-13	max
chess puzzles	Epoch AI	0.2±0.04	2026-10-13	low
frontiermath erdos	Epoch AI	0.0±0.00	2026-08-28	max
frontiermath tier 4 v2	Epoch AI	0.4±0.08	2026-07-17	max
frontiermath tiers 1 3 v2	Epoch AI	0.7±0.03	2026-07-17	max
furniture assembly	Epoch AI	0.3±0.06	2026-09-11	max
gpqa diamond	Epoch AI	0.8±0.03	2026-08-07	low
gpqa diamond	Epoch AI	0.9±0.02	2026-08-07	high
gpqa diamond	Epoch AI	0.9±0.01	2026-07-16	max
mirrorcode	Epoch AI	0.5±0.11	2026-09-03	max
mystery game puzzles	Epoch AI	0.1±0.03	2026-08-29	low
mystery game puzzles	Epoch AI	0.3±0.04	2026-07-29	max
otis mock aime 2024 2025near its ceiling	Epoch AI	0.9±0.04	2026-08-07	high
otis mock aime 2024 2025near its ceiling	Epoch AI	0.7±0.07	2026-08-07	low
otis mock aime 2024 2025near its ceiling	Epoch AI	1.0±0.01	2026-07-16	max
simpleqa verified	Epoch AI	0.5±0.02	2026-08-27	max
ale bench	collected by Epoch AI	1524	—	max
apex agents	collected by Epoch AI	0.6	—	unknown
arc aginear its ceiling	collected by Epoch AI	0.9	—	max
arc aginear its ceiling	collected by Epoch AI	0.7	—	low
arc aginear its ceiling	collected by Epoch AI	0.9	—	high
arc agi 2	collected by Epoch AI	0.1	—	low
arc agi 2	collected by Epoch AI	0.6	—	high
arc agi 2	collected by Epoch AI	0.6	—	max
blueprint bench 2	collected by Epoch AI	0.3	—	unknown
critpt	collected by Epoch AI	0.2	—	max
critpt	collected by Epoch AI	0.0	—	low
deepswe	collected by Epoch AI	0.7	—	mini-swe-agent
dtbenchnear its ceiling	collected by Epoch AI	0.9	—	max
forecastbench	collected by Epoch AI	64.2	—	max
frontiercode	collected by Epoch AI	0.4	—	mini-swe-agent
frontierswe	collected by Epoch AI	0.3	—	proximus
gbaeval	collected by Epoch AI	0.5	—	unknown
gdp pdf	collected by Epoch AI	0.2	—	max
lmca	collected by Epoch AI	52.7	—	max
posttrainbench	collected by Epoch AI	0.3	—	Claude Code
proofbench	collected by Epoch AI	0.9	—	unknown
scicode	collected by Epoch AI	0.5	—	low
scicode	collected by Epoch AI	0.6	—	unknown
scicode	collected by Epoch AI	0.6	—	max
simplebench	collected by Epoch AI	0.6	—	max
surface evolver bench	collected by Epoch AI	0.9	—	unknown
surface evolver bench	collected by Epoch AI	0.9	—	max
vending bench 2	collected by Epoch AI	5165	—	unknown
webdev arena	collected by Epoch AI	1674	—	max
weirdml	collected by Epoch AI	0.8	—	max
weirdml v3	collected by Epoch AI	0.1	—	kimi_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.

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