ZMIME
Comparison · 3 models · Updated Nov 22, 2026

GPT-5.4 vs GPT-6 Astra vs Kimi K3

Too close to call on our weighted score (GPT-6 Astra 66, Kimi K3 65, GPT-5.4 65). The right pick depends on what you value most.GPT-6 Astra comes out ahead, 89 to 82 and 82 on our weighted score, though GPT-5.4 is 3.6× cheaper per token.

  1. OpenAI

    GPT-5.4

    Released Mar 5, 2026

    65/100
    • ECI157.9
    • Price$2.50 / $15.00
    • Context1.05M
  2. OpenAI

    GPT-6 Astra

    Released Sep 4, 2026

    66/100
    • ECI169.2
    • Price$10.00 / $50.00
    • Context1.05M
  3. Moonshot AI

    Kimi K3

    Released Jul 16, 2026Latest

    65/100
    • ECI158.6
    • Price$3.00 / $15.00
    • Context1.05M
  4. OpenAI

    GPT-5.4

    Released Mar 5, 2026

    82/100
    • ECI157.9
    • Price$2.50 / $15.00
    • Context1.05M
  5. Our pick

    OpenAI

    GPT-6 Astra

    Released Sep 4, 2026

    89/100
    • ECI169.2
    • Price$10.00 / $50.00
    • Context1.05M
  6. Moonshot AI

    Kimi K3

    Released Jul 16, 2026Latest

    82/100
    • ECI158.6
    • Price$3.00 / $15.00
    • Context1.05M
01 — Verdict

Too close to callGPT-6 Astra is our pick

It is close. Our weighted score puts them within 2 points (GPT-6 Astra 66/100, Kimi K3 65/100, GPT-5.4 65/100), so choose by what matters most for your work: GPT-6 Astra for raw capability and GPT-5.4 on price. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.GPT-6 Astra is the better all-round choice, scoring 89/100 against Kimi K3 (82) and GPT-5.4 (82). It leads on capability. The score weighs capability 67%, inputs & features 20%, context window 13%.

  • CapabilityGPT-6 AstraCapabilities Index (ECI): GPT-6 Astra 169.2 · Kimi K3 158.6 · GPT-5.4 157.9
  • Lowest priceGPT-5.4GPT-5.4 $5.63 · Kimi K3 $6.00 · GPT-6 Astra $20.00 per 1M tokens (3:1 blend)
  • Longest contextAbout the sameGPT-5.4 1,050,000 · GPT-6 Astra 1,050,000 · Kimi K3 1,048,576 tokens
  • Widest inputsSame inputsGPT-5.4: Text, Images, PDFs · GPT-6 Astra: Text, Images, PDFs · Kimi K3: Text, Images, Video
  • Self-hostingKimi K3Publishes downloadable weights
How the score is built
MeasureWeightGPT-5.4GPT-6 AstraKimi K3
CapabilityCapabilities Index (ECI)50%859686
Price25%14012
Inputs & features15%808080
Context window10%656565
Overall100%65/10066/10065/100
How the score is built
MeasureWeightGPT-5.4GPT-6 AstraKimi K3
CapabilityCapabilities Index (ECI)67%859686
Inputs & features20%808080
Context window13%656565
Overall100%82/10089/10082/100
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.

GPT-5.4 vs GPT-6 Astra vs Kimi K3 specifications side by side
SpecificationGPT-5.4OpenAIGPT-6 AstraOpenAIKimi K3Moonshot AI
Capability
Capabilities Index (ECI)157.9169.2 (best)158.6
ECI rank#29 of 159#4 of 159 (best)#26 of 159
ECI confidence interval156.0–160.0166.4–172.5156.7–160.8
ECI evaluation dateMar 5, 2026Sep 3, 2026Jul 16, 2026
Documented capabilitiesAgent-ready GPT for coding and computer-use workflows at a lower costGPT-6 Astra is OpenAI's most capable model for complex reasoning, coding, computer use, research, and document creation.Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work
Model parametersNot publishedNot publishedNot published
Output modalitiesTextTextText
GPQA DiamondGraduate-level science questions93.3%xhigh95.8%max (best)93.1%max
FrontierMath Tiers 1–3Research-level mathematics78.6%xhigh93.7%max (best)72.2%max
OTIS Mock AIME 2024–2025Competition mathematics97.8%high100%max (best)97.2%max
SWE-bench VerifiedFixing real GitHub issues76.9%highNot publishedNot published
SimpleQA VerifiedShort factual questions45.1%xhigh75.6%max (best)50.6%max
Publisher-reported benchmarks
SWE-Bench Pro · publicPublisher-reported result59.1resolve rateReported resultNot publishedNot published
SWE-Atlas Codebase QnAPublisher-reported result40.8scoreReported resultNot publishedNot published
SWE-Atlas RefactoringPublisher-reported result44.29scoreReported resultNot publishedNot published
SWE-Atlas Test WritingPublisher-reported result44.36scoreReported resultNot publishedNot published
Artificial Analysis Coding Agent Indexmedium53.6average pass@1Reported resultNot publishedNot published
SWE-Atlas Codebase QnAmedium72.4pass@1Reported resultNot publishedNot published
SWE-Bench Pro · hard-aamedium18.4pass@1Reported resultNot publishedNot published
Terminal-Bench 2.1medium69.8pass@1Reported resultNot publishedNot published
Terminal-Bench 2.0Publisher-reported result75.1success rateReported resultApr 23, 2026Not publishedNot published
GPQA DiamondPublisher-reported result92.8accuracyReported resultApr 23, 202696accuracyReported resultSep 3, 2026Not published
Humanity's Last Examno tools39.8accuracyReported resultApr 23, 2026Not publishedNot published
Humanity's Last Examwith tools52.1accuracyReported resultApr 23, 202657.2accuracyReported resultSep 3, 2026Not published
OSWorld-VerifiedPublisher-reported result75success rateReported resultApr 23, 2026Not publishedNot published
BrowseCompPublisher-reported result82.7accuracyReported resultApr 23, 202691.5accuracyReported resultSep 3, 2026Not published
GDPvalPublisher-reported result83wins or tiesReported resultApr 23, 2026Not publishedNot published
ARC-AGI-2Verified73.3accuracyReported resultApr 23, 2026Not publishedNot published
FrontierMath · Tier 1-3Publisher-reported result47.6accuracyReported resultApr 23, 2026Not publishedNot published
FrontierMath · Tier 4Publisher-reported result27.1accuracyReported resultApr 23, 2026Not publishedNot published
MMMU Prono tools81.2accuracyReported resultApr 23, 2026Not publishedNot published
Agents' Last ExamPublisher-reported resultNot published59.3Reported resultSep 3, 2026Not published
OSWorld 2.0 · V2-Offline (v2026.08.08)Publisher-reported resultNot published72.6partial scoreReported resultSep 3, 2026Not published
ScreenSpot-Prono toolsNot published92.7accuracyReported resultSep 3, 2026Not published
AutomationBenchPublisher-reported resultNot published41.4success rateReported resultSep 3, 2026Not published
BenchCADwith toolsNot published95.9geometric overlapReported resultSep 3, 2026Not published
Terminal-Bench 4.0Publisher-reported resultNot published57.9success rateReported resultSep 3, 2026Not published
DeepSWE 1.1Publisher-reported resultNot published74.1resolve rateReported resultSep 3, 2026Not published
FrontierCode 1.1 · ExtendedPublisher-reported resultNot published64.5scoreReported resultSep 3, 2026Not published
Terminal-Bench Science 0.1Publisher-reported resultNot published64.6success rateReported resultSep 3, 2026Not published
FrontierMath v2 · Tier 4Publisher-reported resultNot published97.6accuracyReported resultSep 3, 2026Not published
ExploitBenchwithout production safeguardsNot published100success rateReported resultSep 3, 2026Not published
SRE-Benchsingle attemptNot published88success rateReported resultSep 3, 2026Not published
ARC-AGI-3Publisher-reported resultNot published99.9RHAEReported resultSep 3, 2026Not published
Artificial Analysis Intelligence Index 4.1.1maxNot published61index scoreReported resultSep 3, 2026Not published
Artificial Analysis Coding Agent Index 1.4maxNot published67index scoreReported resultSep 3, 2026Not published
DeepSWE 1.1max effortNot publishedNot published67.5resolve rateReported resultJul 16, 2026
Terminal-Bench 2.1max effortNot publishedNot published88.3accuracyReported resultJul 16, 2026
FrontierSWEmax effortNot publishedNot published81.2dominance scoreReported resultJul 16, 2026
Program Benchmax effortNot publishedNot published77.8scoreReported resultJul 16, 2026
SWE Marathon 1.1max effortNot publishedNot published42resolve rateReported resultJul 16, 2026
GDPval-AA v2max effortNot publishedNot published1668EloReported resultJul 16, 2026
AA-Briefcasemax effortNot publishedNot published1548EloReported resultJul 16, 2026
AutomationBench · 600-task public subsetmax effortNot publishedNot published30.8success rateReported resultJul 16, 2026
JobBenchmax effortNot publishedNot published52.9scoreReported resultJul 16, 2026
SpreadsheetBench 2max effortNot publishedNot published34.8scoreReported resultJul 16, 2026
BrowseCompmax effort, context compactionNot publishedNot published91.2accuracyReported resultJul 16, 2026
CharXiv Reasoningmax effort, with toolsNot publishedNot published91.3accuracyReported resultJul 16, 2026
ZeroBenchmax effort, with toolsNot publishedNot published41pass@5Reported resultJul 16, 2026
Price per million tokens
Input$2.50 (best)$10.00$3.00
Output$15.00 (best)$50.00$15.00 (best)
Cached input$0.25 (best)$1.00$0.30
Blended (3:1)$5.63 (best)$20.00$6.00
Long-context rateOver 272K: $5.00 / $22.50Over 272K: $20.00 / $75.00Same rate
Price sourceOfficial OpenAI APIOfficial OpenAI APIOfficial Moonshot AI API
Limits
Context window1,050,000 tokens (best)1,050,000 tokens (best)1,048,576 tokens
Max input922,000 tokens922,000 tokensNot published
Max output128,000 tokens128,000 tokens1,048,576 tokens (best)
Inputs and features
TextYesYesYes
ImagesYesYesYes
PDFsYesYesNo
AudioNoNoNo
VideoNoNoYes
ReasoningYeslow · medium · high · xhighYeslow · medium · high · xhigh · maxYeslow · high · max
Tool callingYesYesYes
Structured outputYesYesYes
File attachmentsYesYesYes
Temperature controlYesNoNo
Reasoning controleffortefforteffort
Availability
WeightsProprietaryProprietaryOpen
API model IDgpt-5.4gpt-6-astrakimi-k3
API providers505693 (best)
ReleasedMar 5, 2026Sep 4, 2026Jul 16, 2026
Knowledge cutoffAug 31, 2025Apr 30, 2026Not published
Publisher documentationPublisher documentationPublisher documentationPublisher documentation · Model weights
Source metadata updatedMar 5, 2026Sep 4, 2026Jul 16, 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.

  • GPT-5.4$55.00
  • GPT-6 Astra$200.00
  • Kimi K3$60.00
04 — Questions

Which should you choose?

Which is better: GPT-5.4, GPT-6 Astra or Kimi K3?

It is close. Our weighted score puts them within 2 points (GPT-6 Astra 66/100, Kimi K3 65/100, GPT-5.4 65/100), so choose by what matters most for your work: GPT-6 Astra for raw capability and GPT-5.4 on price. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.GPT-6 Astra is the better all-round choice, scoring 89/100 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, GPT-5.4, GPT-6 Astra or Kimi K3?

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); GPT-6 Astra costs $10.00 input / $50.00 output per million tokens (official OpenAI 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 GPT-6 Astra (3.6× as much).

Which scores higher on benchmarks?

GPT-6 Astra scores higher on the Capabilities Index (ECI): GPT-6 Astra 169.2 (#4 of 159), Kimi K3 158.6 (#26 of 159) and GPT-5.4 157.9 (#29 of 159). Their confidence ranges do not overlap (166.4–172.5 vs 156.7–160.8), so the gap is a real one. On individual benchmarks: GPQA Diamond — GPT-6 Astra 95.8%, GPT-5.4 93.3%, Kimi K3 93.1%; FrontierMath Tiers 1–3 — GPT-6 Astra 93.7%, GPT-5.4 78.6%, Kimi K3 72.2%; OTIS Mock AIME 2024–2025 — GPT-6 Astra 100%, GPT-5.4 97.8%, Kimi K3 97.2%; SimpleQA Verified — GPT-6 Astra 75.6%, Kimi K3 50.6%, GPT-5.4 45.1%.

Which is better for coding?

There are no published SWE-bench Verified results for GPT-6 Astra and Kimi K3 yet, so there is no like-for-like coding score. On overall capability, GPT-6 Astra 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?

Their context windows are effectively the same size: GPT-5.4 1,050,000, GPT-6 Astra 1,050,000 and Kimi K3 1,048,576 tokens. Maximum output per response: GPT-5.4 up to 128,000, GPT-6 Astra up to 128,000, Kimi K3 up to 1,048,576 tokens.

Which can read images, PDFs, audio or video?

GPT-5.4 accepts text, images and PDFs; GPT-6 Astra accepts text, images and PDFs; Kimi K3 accepts text, images and video. They handle the same number of input types.

Are any of these open source?

Kimi K3 publishes its weights and can be self-hosted; GPT-5.4 and GPT-6 Astra is proprietary.

Which is newer?

GPT-6 Astra is the newest, released Sep 4, 2026. Kimi K3 came out Jul 16, 2026; GPT-5.4 came out Mar 5, 2026. Knowledge cutoff: GPT-5.4 Aug 31, 2025, GPT-6 Astra Apr 30, 2026.

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.