ZMIME
Comparison · 3 models · Updated Nov 22, 2026

Kimi K3 vs Claude Fable 5.1 vs GPT-5.4

Too close to call on our weighted score (Claude Fable 5.1 65, Kimi K3 65, GPT-5.4 65). The right pick depends on what you value most.Claude Fable 5.1 comes out ahead, 87 to 82 and 82 on our weighted score, though GPT-5.4 is 3.6× cheaper per token.

  1. Moonshot AI

    Kimi K3

    Released Jul 16, 2026Latest

    65/100
    • ECI158.6
    • Price$3.00 / $15.00
    • Context1.05M
  2. Anthropic

    Claude Fable 5.1

    Released Sep 1, 2026

    65/100
    • ECI167.1
    • Price$10.00 / $50.00
    • Context1M
  3. OpenAI

    GPT-5.4

    Released Mar 5, 2026

    65/100
    • ECI157.9
    • Price$2.50 / $15.00
    • Context1.05M
  4. Moonshot AI

    Kimi K3

    Released Jul 16, 2026Latest

    82/100
    • ECI158.6
    • Price$3.00 / $15.00
    • Context1.05M
  5. Our pick

    Anthropic

    Claude Fable 5.1

    Released Sep 1, 2026

    87/100
    • ECI167.1
    • Price$10.00 / $50.00
    • Context1M
  6. OpenAI

    GPT-5.4

    Released Mar 5, 2026

    82/100
    • ECI157.9
    • Price$2.50 / $15.00
    • Context1.05M
01 — Verdict

Too close to callClaude Fable 5.1 is our pick

It is close. Our weighted score puts them within a point (Claude Fable 5.1 65/100, Kimi K3 65/100, GPT-5.4 65/100), so choose by what matters most for your work: Claude Fable 5.1 for raw capability and GPT-5.4 on price. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.Claude Fable 5.1 is the better all-round choice, scoring 87/100 against Kimi K3 (82) and GPT-5.4 (82). It leads on capability. The score weighs capability 67%, inputs & features 20%, context window 13%.

  • CapabilityClaude Fable 5.1Capabilities Index (ECI): Claude Fable 5.1 167.1 · Kimi K3 158.6 · GPT-5.4 157.9
  • Lowest priceGPT-5.4GPT-5.4 $5.63 · Kimi K3 $6.00 · Claude Fable 5.1 $20.00 per 1M tokens (3:1 blend)
  • Longest contextGPT-5.4 and Kimi K3GPT-5.4 1,050,000 · Kimi K3 1,048,576 · Claude Fable 5.1 1,000,000 tokens
  • Widest inputsSame inputsKimi K3: Text, Images, Video · Claude Fable 5.1: Text, Images, PDFs · GPT-5.4: Text, Images, PDFs
  • Self-hostingKimi K3Publishes downloadable weights
How the score is built
MeasureWeightKimi K3Claude Fable 5.1GPT-5.4
CapabilityCapabilities Index (ECI)50%869485
Price25%12014
Inputs & features15%808080
Context window10%656465
Overall100%65/10065/10065/100
How the score is built
MeasureWeightKimi K3Claude Fable 5.1GPT-5.4
CapabilityCapabilities Index (ECI)67%869485
Inputs & features20%808080
Context window13%656465
Overall100%82/10087/10082/100
02 — Side by side

Every spec in one table

Highlighted cells lead their row. Unavailable specifications are marked “Not published”. Publisher results include their evaluation settings and source.

Kimi K3 vs Claude Fable 5.1 vs GPT-5.4 specifications side by side
SpecificationKimi K3Moonshot AIClaude Fable 5.1AnthropicGPT-5.4OpenAI
Capability
Capabilities Index (ECI)158.6167.1 (best)157.9
ECI rank#26 of 159#8 of 159 (best)#29 of 159
ECI confidence interval156.7–160.8164.4–170.1156.0–160.0
ECI evaluation dateJul 16, 2026Sep 1, 2026Mar 5, 2026
Documented capabilitiesMultimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent workClaude model for demanding reasoning and long-horizon agentic workAgent-ready GPT for coding and computer-use workflows at a lower cost
Model parametersNot publishedNot publishedNot published
Output modalitiesTextTextText
GPQA DiamondGraduate-level science questions93.1%maxNot published93.3%xhigh (best)
FrontierMath Tiers 1–3Research-level mathematics72.2%max90.2%max (best)78.6%xhigh
OTIS Mock AIME 2024–2025Competition mathematics97.2%max100%max (best)97.8%high
SWE-bench VerifiedFixing real GitHub issuesNot publishedNot published76.9%high
SimpleQA VerifiedShort factual questions50.6%max70.8%max (best)45.1%xhigh
Publisher-reported benchmarks
DeepSWE 1.1max effort67.5resolve rateReported resultJul 16, 2026Not publishedNot published
Terminal-Bench 2.1max effort88.3accuracyReported resultJul 16, 2026Not publishedNot published
FrontierSWEmax effort81.2dominance scoreReported resultJul 16, 2026Not publishedNot published
Program Benchmax effort77.8scoreReported resultJul 16, 2026Not publishedNot published
SWE Marathon 1.1max effort42resolve rateReported resultJul 16, 2026Not publishedNot published
GDPval-AA v2max effort1668EloReported resultJul 16, 2026Not publishedNot published
AA-Briefcasemax effort1548EloReported resultJul 16, 2026Not publishedNot published
AutomationBench · 600-task public subsetmax effort30.8success rateReported resultJul 16, 2026Not publishedNot published
JobBenchmax effort52.9scoreReported resultJul 16, 2026Not publishedNot published
SpreadsheetBench 2max effort34.8scoreReported resultJul 16, 2026Not publishedNot published
BrowseCompmax effort, context compaction91.2accuracyReported resultJul 16, 2026Not publishedNot published
CharXiv Reasoningmax effort, with tools91.3accuracyReported resultJul 16, 2026Not publishedNot published
ZeroBenchmax effort, with tools41pass@5Reported resultJul 16, 2026Not publishedNot published
Terminal-Bench-Science 0.1production safeguards with fallbackNot published52.6scoreReported resultSep 1, 2026Not published
Terminal-Bench 4.0production safeguards with fallbackNot published55.8scoreReported resultSep 1, 2026Not published
GDPval-AA 2production safeguards with fallbackNot published1853EloReported resultSep 1, 2026Not published
OSWorld 2.0 · August 2026 task releaseproduction safeguards; interventions score zero; no fallbackNot published77.9partial scoreReported resultSep 1, 2026Not published
OSWorld 2.0 · August 2026 task releaseproduction safeguards; interventions score zero; no fallbackNot published41.7strict success rateReported resultSep 1, 2026Not published
Humanity's Last Examwithout tools; production safeguards with fallbackNot published60.9scoreReported resultSep 1, 2026Not published
Humanity's Last Examwith tools; production safeguards with fallbackNot published65scoreReported resultSep 1, 2026Not published
AutomationBenchproduction safeguards with fallbackNot published31.4pass rateReported resultSep 1, 2026Not published
CursorBench 3.2.0max effort; production safeguards with fallbackNot published73.4scoreReported resultSep 1, 2026Not published
SWE-Bench Pro · publicPublisher-reported resultNot publishedNot published59.1resolve rateReported result
SWE-Atlas Codebase QnAPublisher-reported resultNot publishedNot published40.8scoreReported result
SWE-Atlas RefactoringPublisher-reported resultNot publishedNot published44.29scoreReported result
SWE-Atlas Test WritingPublisher-reported resultNot publishedNot published44.36scoreReported result
Artificial Analysis Coding Agent IndexmediumNot publishedNot published53.6average pass@1Reported result
SWE-Atlas Codebase QnAmediumNot publishedNot published72.4pass@1Reported result
SWE-Bench Pro · hard-aamediumNot publishedNot published18.4pass@1Reported result
Terminal-Bench 2.1mediumNot publishedNot published69.8pass@1Reported result
Terminal-Bench 2.0Publisher-reported resultNot publishedNot published75.1success rateReported resultApr 23, 2026
GPQA DiamondPublisher-reported resultNot publishedNot published92.8accuracyReported resultApr 23, 2026
Humanity's Last Examno toolsNot publishedNot published39.8accuracyReported resultApr 23, 2026
Humanity's Last Examwith toolsNot publishedNot published52.1accuracyReported resultApr 23, 2026
OSWorld-VerifiedPublisher-reported resultNot publishedNot published75success rateReported resultApr 23, 2026
BrowseCompPublisher-reported resultNot publishedNot published82.7accuracyReported resultApr 23, 2026
GDPvalPublisher-reported resultNot publishedNot published83wins or tiesReported resultApr 23, 2026
ARC-AGI-2VerifiedNot publishedNot published73.3accuracyReported resultApr 23, 2026
FrontierMath · Tier 1-3Publisher-reported resultNot publishedNot published47.6accuracyReported resultApr 23, 2026
FrontierMath · Tier 4Publisher-reported resultNot publishedNot published27.1accuracyReported resultApr 23, 2026
MMMU Prono toolsNot publishedNot published81.2accuracyReported resultApr 23, 2026
Price per million tokens
Input$3.00$10.00$2.50 (best)
Output$15.00 (best)$50.00$15.00 (best)
Cached input$0.30$0.25 (best)$0.25 (best)
Blended (3:1)$6.00$20.00$5.63 (best)
Long-context rateSame rateSame rateOver 272K: $5.00 / $22.50
Price sourceOfficial Moonshot AI APIOfficial Anthropic APIOfficial OpenAI API
Limits
Context window1,048,576 tokens1,000,000 tokens1,050,000 tokens (best)
Max inputNot publishedNot published922,000 tokens
Max output1,048,576 tokens (best)128,000 tokens128,000 tokens
Inputs and features
TextYesYesYes
ImagesYesYesYes
PDFsNoYesYes
AudioNoNoNo
VideoYesNoNo
ReasoningYeslow · high · maxYeslow · medium · high · xhigh · maxYeslow · medium · high · xhigh
Tool callingYesYesYes
Structured outputYesYesYes
File attachmentsYesYesYes
Temperature controlNoNoYes
Reasoning controleffortefforteffort
Availability
WeightsOpenProprietaryProprietary
API model IDkimi-k3claude-fable-5-1gpt-5.4
API providers93 (best)5150
ReleasedJul 16, 2026Sep 1, 2026Mar 5, 2026
Knowledge cutoffNot publishedJun 2026Aug 31, 2025
Publisher documentationPublisher documentation · Model weightsPublisher documentationPublisher documentation
Source metadata updatedJul 16, 2026Sep 1, 2026Mar 5, 2026
Data sourcesmodels.dev · Epoch AImodels.dev · Epoch AImodels.dev · Epoch AI
Catalog checkedNov 22, 2026Nov 22, 2026Nov 22, 2026
03 — Cost

What would a month cost?

Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.

  • Kimi K3$60.00
  • Claude Fable 5.1$200.00
  • GPT-5.4$55.00
04 — Questions

Which should you choose?

Which is better: Kimi K3, Claude Fable 5.1 or GPT-5.4?

It is close. Our weighted score puts them within a point (Claude Fable 5.1 65/100, Kimi K3 65/100, GPT-5.4 65/100), so choose by what matters most for your work: Claude Fable 5.1 for raw capability and GPT-5.4 on price. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.Claude Fable 5.1 is the better all-round choice, scoring 87/100 against Kimi K3 (82) and GPT-5.4 (82). It leads on capability. The score weighs capability 67%, inputs & features 20%, context window 13%.

Which is cheaper, Kimi K3, Claude Fable 5.1 or GPT-5.4?

GPT-5.4 is cheaper at $2.50 input / $15.00 output per million tokens (official OpenAI API price). Kimi K3 costs $3.00 input / $15.00 output per million tokens (official Moonshot AI API price); Claude Fable 5.1 costs $10.00 input / $50.00 output per million tokens (official Anthropic API price). At a typical mix of three input tokens to one output token, that is $5.63 per million tokens for GPT-5.4 versus $6.00 for Kimi K3 (1.1× as much) and $20.00 for Claude Fable 5.1 (3.6× as much).

Which scores higher on benchmarks?

Claude Fable 5.1 scores higher on the Capabilities Index (ECI): Claude Fable 5.1 167.1 (#8 of 159), Kimi K3 158.6 (#26 of 159) and GPT-5.4 157.9 (#29 of 159). Their confidence ranges do not overlap (164.4–170.1 vs 156.7–160.8), so the gap is a real one. On individual benchmarks: FrontierMath Tiers 1–3 — Claude Fable 5.1 90.2%, GPT-5.4 78.6%, Kimi K3 72.2%; OTIS Mock AIME 2024–2025 — Claude Fable 5.1 100%, GPT-5.4 97.8%, Kimi K3 97.2%; SimpleQA Verified — Claude Fable 5.1 70.8%, Kimi K3 50.6%, GPT-5.4 45.1%.

Which is better for coding?

There are no published SWE-bench Verified results for Kimi K3 and Claude Fable 5.1 yet, so there is no like-for-like coding score. On overall capability, Claude Fable 5.1 leads, which tends to carry over to coding, but test on your own codebase. All three support tool calling for agent workflows.

Which has the bigger context window?

GPT-5.4 and Kimi K3 have the largest context windows (1,050,000 and 1,048,576 tokens), against 1,000,000 for Claude Fable 5.1. Maximum output per response: Kimi K3 up to 1,048,576, Claude Fable 5.1 up to 128,000, GPT-5.4 up to 128,000 tokens.

Which can read images, PDFs, audio or video?

Kimi K3 accepts text, images and video; Claude Fable 5.1 accepts text, images and PDFs; GPT-5.4 accepts text, images and PDFs. They handle the same number of input types.

Are any of these open source?

Kimi K3 publishes its weights and can be self-hosted; Claude Fable 5.1 and GPT-5.4 is proprietary.

Which is newer?

Claude Fable 5.1 is the newest, released Sep 1, 2026. Kimi K3 came out Jul 16, 2026; GPT-5.4 came out Mar 5, 2026. Knowledge cutoff: Claude Fable 5.1 Jun 2026, GPT-5.4 Aug 31, 2025.

How do you decide the winner?

This rating uses capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.This rating uses capability (67%, independent benchmark results); inputs & features (20%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (13%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 22, 2026.