priorsBuilder
Support

Settings

Theme

Explore themes

Mode

Library

Looks borrowed from benchmark sites and products. Themes with a light and a dark version follow your mode.

Bring your own theme

Upload a Priors theme file, or have your agent make one: copy the prompt, tell it the look you want, and upload what it returns. With the Priors skill installed it already knows the format.

Custom themes stay in this browser. They travel inside shared chart links and a page's theme, so others see them as you do.

For the builder, Explore and model pages on this device. Published pages keep the theme their owner picked.

Mode

Moonshot · released 2026-07-16

Kimi K3

Measured on 238 benchmarks across 2 reasoning efforts.

29.8

#25 of 63 models · best at max

Reasoning efforts in this ranking

max29.8#68 of 146
low17.4#114 of 146

Strengths

  • Kimi K3
  • Gemini 4 Carbon (#1)
  • Field median
bestField medianFrontierCode 2.0-…Terminal-Bench v4…AA-Briefcase v1.1AutomationBench-AATerminal-Bench-Sc…HLE-DiamondCritPt v2 Hard Su…AA-Omniscience v1…AA-Omniscience v1…GDP.pdf
  1. FrontierCode 2.0-AA15%10.9% est.#38 of 63
  2. Terminal-Bench v4-AA10%12.6%#36 of 63
  3. AA-Briefcase v1.115%1497#21 of 63
  4. AutomationBench-AA5%58.3%#25 of 63
  5. Terminal-Bench-Science-AA15%8.1%#25 of 63
  6. HLE-Diamond10%26.9%#20 of 63
  7. CritPt v2 Hard Subset10%40.5%#20 of 63
  8. AA-Omniscience v1.1 Accuracy10%42.4% est.#8 of 63
  9. AA-Omniscience v1.1 Non-hallucination5%23.4% est.#52 of 63
  10. GDP.pdf5%22.0%#18 of 63

Bars show where Kimi K3 sits among the 63 ranked models on each benchmark (right is best); the tick is the field median.

Cost and speed

against 63 ranked models

Cost per task

#21 cheapest of 44

$0.18at low

Time per task

#25 quickest of 43

17 minat low

Output speed

#49 fastest of 54

44 tok/sat max

Time to first token

#31 quickest of 54

7.3sat max

Output tokens per task

#7 fewest of 47

40kat low

Token price

$0.441 per 1M tokens at API prices$0.034 on Kimi Ultra

97.0% cache read · 2.50% fresh input · 0.00% cache write · 0.50% output

Assumed mix: no measured T3 Code usage for this model.

This ranking’s benchmarks

10 of 10 measured

agentic

AutomationBench-AA

241 results
  1. #51max of 24158.3%
  2. #94low of 24149.3%

Leader: Gemini 4 Carbon · 77.8%

agentic

AA-Briefcase v1.1

247 results
  1. #40max of 2471497
  2. #142low of 247985

Leader: Sonnet 5.5 · 1813

reasoning

CritPt v2 Hard Subset

111 results
  1. #47max of 11140.5%
  2. #91low of 1115.5%

Leader: GPT-6 Astra · 72.0%

coding

FrontierCode 2.0-AA

108 results
  1. #79low of 10810.9%

Leader: Opus 5.5 · 46.4%

economics

GDP.pdf

240 results
  1. #60max of 24022.0%
  2. #85low of 24018.6%

Leader: Muse Wonder Full · 32.6%

reasoning

HLE-Diamond

145 results
  1. #45max of 14526.9%
  2. #87low of 14514.0%

Leader: Gemini 4 Carbon · 69.1%

knowledge

AA-Omniscience v1.1 Accuracy

117 results
  1. #46low of 11742.4%

Leader: Fable 5.1 · 68.4%

knowledge

AA-Omniscience v1.1 Non-hallucination

117 results
  1. #90low of 11723.4%

Leader: Gemini 4 Carbon · 95.8%

reasoning

Terminal-Bench-Science-AA

120 results
  1. #60max of 1208.1%
  2. #89low of 1201.0%

Leader: Gemini 4 Carbon · 64.8%

coding

Terminal-Bench v4-AA

238 results
  1. #95low of 23812.6%
  2. #97max of 23812.6%

Leader: Kimi K3.1 · 67.2%

All 238 benchmarks

show 228 more