Decision support

Claude Fable 5.2 vs Kimi K3

Compare canonical objects directly. The useful dimensions change with the object type, so model and hardware comparisons stay separate.

Models Hardware

Claude Fable 5.2

Anthropic

Sources:models.devOpenRouter

Kimi K3

Moonshot AI

Sources:models.devOpenRouter

Benchmark results

Head to head on 8 benchmarks both models report

Claude Fable 5.28leads
Kimi K30leads

DeepSWE 1.1
Best pass@1 · Higher is better
- Claude Fable 5.2
  Not reported
- Kimi K3
  68.5%
  #8 of 28
Runs and sources
Claude Fable 5.2
Not reported
Kimi K3
max68.5%mini-swe-agent · pass@4 89.4% · $4.65 / attempt
Source:DeepSWE leaderboard, 2026-09-22 Official source JSON
Arena Text
Arena rating · Higher is better
- Claude Fable 5.2
  Not reported
- Kimi K3
  1487
  Arena #19CI 1483 to 1491
Runs and sources
Claude Fable 5.2
Not reported
Kimi K3
max148795% CI 1483 to 1491 · 40,867 votes
Source:Arena Text leaderboard, 2026-11-09
Arena WebDev
Arena rating · Higher is better
- Claude Fable 5.2
  Best1836
  +178 ahead · Arena #1CI 1813 to 1858
- Kimi K3
  1658
  Arena #18CI 1652 to 1664
Runs and sources
Claude Fable 5.2
xhigh183695% CI 1813 to 1858 · 994 votes
Kimi K3
max165895% CI 1652 to 1664 · 22,405 votes
Source:Arena WebDev leaderboard, 2026-11-10
Agent Arena
IPS score · Higher is better
- Claude Fable 5.2
  Best+0.304
  +0.298 ahead · Arena #1CI +0.264 to +0.343
- Kimi K3
  +0.006
  Arena #28CI -0.001 to +0.012
Runs and sources
Claude Fable 5.2
max+0.30495% CI +0.264 to +0.343 · 577,927 observations
Kimi K3
max+0.00695% CI -0.001 to +0.012 · 16,124,377 observations
Source:Agent Arena leaderboard, 2026-11-11
ARC-AGI-2
Best published score · Higher is better
- Claude Fable 5.2
  Best93.8%
  +33.3 pts ahead · #4 of 85
- Kimi K3
  60.4%
  #36 of 85
Runs, sources and effort chart
Claude Fable 5.2
low86.7%
medium89.6%
high90.4%
xhigh93.8%
max92.5%
Kimi K3
low12.4%
high55.0%
max60.4%
Source:External evaluation via Epoch AIOriginal-source terms apply
FrontierMath-Tier-4-v2-Private
Best published score · Higher is better
- Claude Fable 5.2
  Best100.0%
  +61.0 pts ahead · #1 of 72
- Kimi K3
  39.0%
  #28 of 72
Runs and sources
Claude Fable 5.2
max100.0%
Kimi K3
max39.0%
Source:Epoch AI evaluation
FrontierMath-Tiers-1-3-v2-Private
Best published score · Higher is better
- Claude Fable 5.2
  Best95.1%
  +22.9 pts ahead · #1 of 87
- Kimi K3
  72.2%
  #26 of 87
Runs and sources
Claude Fable 5.2
max95.1%
Kimi K3
max72.2%
Source:Epoch AI evaluation
GPQA diamond
Best published score · Higher is better
- Claude Fable 5.2
  Best93.9%
  +0.8 pts ahead · #14 of 161
- Kimi K3
  93.1%
  #21 of 161
Runs and sources
Claude Fable 5.2
max93.9%
Kimi K3
low84.8%
high91.9%
max93.1%
Source:Epoch AI evaluation
OTIS Mock AIME 2024-2025
Best published score · Higher is better
- Claude Fable 5.2
  Best99.7%
  +2.5 pts ahead · #13 of 158
- Kimi K3
  97.2%
  #32 of 158
Runs and sources
Claude Fable 5.2
max99.7%
Kimi K3
low68.9%
high93.3%
max97.2%
Source:Epoch AI evaluation
SimpleQA Verified
Best published score · Higher is better
- Claude Fable 5.2
  Best94.0%
  +43.4 pts ahead · #1 of 87
- Kimi K3
  50.6%
  #29 of 87
Runs and sources
Claude Fable 5.2
max94.0%
Kimi K3
max50.6%
Source:Epoch AI evaluation

Each model's best published result is shown; "Not reported" means no matched evaluation, not a score of zero. Arena and DeepSWE ranks are the leaderboards' own; other ranks are among catalogue models with a score. Results from different harnesses or settings are not directly comparable.

Original-source terms apply to evaluations credited to their original publishers.

DeepSWE 1.1 results: official leaderboard, data generated 2026-09-22. Only exact model versions in the official dataset are matched.

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Arena ratings: Arena (lmarena-ai/leaderboard-dataset), licensed CC BY 4.0. Ratings come from pairwise human votes and are only comparable within one leaderboard; the interval is Arena's 95% confidence interval. Model names are matched to the Vertical Tabs catalog.

Model specifications

Release date

November 7, 2026

July 16, 2026

Context window

Native context

—

Max output tokens

128,000

943,718

Knowledge cutoff

2026-06

—

Open weights

Closed

Open weights

Parameter count

—

2.8T

API pricing · USD per 1M tokens

Input

$8.00

$0.192

Output

$40.00

$16.00

Cached input

$0.20

$0.182

Cache write

$10.00

—

Capabilities and modalities

Input modalities

text, image, file

text, image, video

Output modalities

text

Tool calling

Supported

Reasoning option

Supported

Supported options

include_reasoning, max_completion_tokens, max_tokens, reasoning, reasoning_effort, response_format, stop, structured_outputs, tool_choice, tools, verbosity

frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p

Capabilities

—

vision, thinking, completion, tools

Architecture and availability

Architecture

—

kimi-k3

Model families

—

kimi-k3

Tokenizer

Claude

Other

Instruction format

—

License

—

Providers

Provider endpoints

Ollama variants

Metadata source

ai-research

A dash means the field is not reported. Provider-specific performance and pricing may differ.

Benchmark results

DeepSWE 1.1

Arena Text

Arena WebDev

Agent Arena

ARC-AGI-2

FrontierMath-Tier-4-v2-Private

FrontierMath-Tiers-1-3-v2-Private

GPQA diamond

OTIS Mock AIME 2024-2025

SimpleQA Verified

Model specifications

API pricing · USD per 1M tokens

Capabilities and modalities

Architecture and availability