ZMIME
Comparison · 3 models · Updated Nov 18, 2026

GPT-5.4 mini vs Kimi K2.5 vs Qwen3.5 397B-A17B

Too close to call on our weighted score (Qwen3.5 397B-A17B 72, GPT-5.4 mini 72, Kimi K2.5 70). The right pick depends on what you value most.

Price in rating: Excluded · Include price

  1. OpenAI

    GPT-5.4 mini

    Released Mar 17, 2026

    72/100
    • ECI149.4
    • Price$0.75 / $4.50
    • Context400K
  2. Moonshot AI

    Kimi K2.5

    Released Jan 27, 2026

    70/100
    • ECI148.8
    • Price$0.60 / $3.00
    • Context262K
  3. Alibaba (Qwen)

    Qwen3.5 397B-A17B

    Released Feb 15, 2026

    72/100
    • ECI148.0
    • Price$0.60 / $3.60
    • Context262K
01 — Verdict

Too close to call

It is close. Our weighted score puts them within a point (Qwen3.5 397B-A17B 72/100, GPT-5.4 mini 72/100, Kimi K2.5 70/100), so choose by what matters most for your work: GPT-5.4 mini for raw capability and Kimi K2.5 on price. The score weighs capability 67%, inputs & features 20%, context window 13%.

  • CapabilityGPT-5.4 miniCapabilities Index (ECI): GPT-5.4 mini 149.4 · Kimi K2.5 148.8 · Qwen3.5 397B-A17B 148.0
  • Lowest priceKimi K2.5Kimi K2.5 $1.20 · Qwen3.5 397B-A17B $1.35 · GPT-5.4 mini $1.69 per 1M tokens (3:1 blend)
  • Longest contextGPT-5.4 miniGPT-5.4 mini 400,000 · Kimi K2.5 262,144 · Qwen3.5 397B-A17B 262,144 tokens
  • Widest inputsKimi K2.5 and Qwen3.5 397B-A17BGPT-5.4 mini: Text, Images · Kimi K2.5: Text, Images, Video · Qwen3.5 397B-A17B: Text, Images, Video
  • Self-hostingKimi K2.5 and Qwen3.5 397B-A17BPublishes downloadable weights
How the score is built
MeasureWeightGPT-5.4 miniKimi K2.5Qwen3.5 397B-A17B
CapabilityCapabilities Index (ECI)67%777675
Inputs & features20%707080
Context window13%504343
Overall100%72/10070/10072/100
02 — Side by side

Every spec in one table

Highlighted cells lead their row. Dashes mean the data is not published.

GPT-5.4 mini vs Kimi K2.5 vs Qwen3.5 397B-A17B specifications side by side
SpecificationGPT-5.4 miniOpenAIKimi K2.5Moonshot AIQwen3.5 397B-A17BAlibaba (Qwen)
Capability
Capabilities Index (ECI)149.4 (best)148.8148.0
ECI rank#71 of 158 (best)#73 of 158#76 of 158
GPQA DiamondGraduate-level science questions86.9%87.6% (best)86.4%
FrontierMath Tiers 1–3Research-level mathematics51.2% (best)—31.2%
OTIS Mock AIME 2024–2025Competition mathematics88.9%92.2% (best)88.9%
SWE-bench VerifiedFixing real GitHub issues—73.8%—
SimpleQA VerifiedShort factual questions29.4%34.3% (best)—
Price per million tokens
Input$0.75$0.60 (best)$0.60 (best)
Output$4.50$3.00 (best)$3.60
Cached input$0.075——
Blended (3:1)$1.69$1.20 (best)$1.35
Long-context rateSame rateSame rateSame rate
Price sourceOfficial OpenAI APIMedian of 36 providersOfficial Alibaba API
Limits
Context window400,000 tokens (best)262,144 tokens262,144 tokens
Max output128,000 tokens262,144 tokens (best)65,536 tokens
Inputs and features
TextYesYesYes
ImagesYesYesYes
PDFsNoNoNo
AudioNoNoNo
VideoNoYesYes
ReasoningYeslow · medium · high · xhighYesYes
Tool callingYesYesYes
Structured outputYesNoYes
Availability
WeightsProprietaryOpenOpen
API model IDgpt-5.4-mini—qwen3.5-397b-a17b
API providers42 (best)3636
ReleasedMar 17, 2026Jan 27, 2026Feb 15, 2026
Knowledge cutoffAug 31, 2025Jan 2025—
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 mini$16.50
  • Kimi K2.5$12.00
  • Qwen3.5 397B-A17B$13.20
04 — Questions

Which should you choose?

Which is better: GPT-5.4 mini, Kimi K2.5 or Qwen3.5 397B-A17B?

It is close. Our weighted score puts them within a point (Qwen3.5 397B-A17B 72/100, GPT-5.4 mini 72/100, Kimi K2.5 70/100), so choose by what matters most for your work: GPT-5.4 mini for raw capability and Kimi K2.5 on price. The score weighs capability 67%, inputs & features 20%, context window 13%.

Which is cheaper, GPT-5.4 mini, Kimi K2.5 or Qwen3.5 397B-A17B?

Kimi K2.5 is cheaper at $0.60 input / $3.00 output per million tokens (median across 36 API providers). Qwen3.5 397B-A17B costs $0.60 input / $3.60 output per million tokens (official Alibaba API price); GPT-5.4 mini costs $0.75 input / $4.50 output per million tokens (official OpenAI API price). At a typical mix of three input tokens to one output token, that is $1.20 per million tokens for Kimi K2.5 versus $1.35 for Qwen3.5 397B-A17B (1.1× as much) and $1.69 for GPT-5.4 mini (1.4× as much).

Which scores higher on benchmarks?

GPT-5.4 mini scores higher on the Capabilities Index (ECI): GPT-5.4 mini 149.4 (#71 of 158), Kimi K2.5 148.8 (#73 of 158) and Qwen3.5 397B-A17B 148.0 (#76 of 158). The confidence ranges of the top two overlap (147.7–151.1 vs 147.3–150.4), so treat the gap as small. On individual benchmarks: GPQA Diamond — Kimi K2.5 87.6%, GPT-5.4 mini 86.9%, Qwen3.5 397B-A17B 86.4%; OTIS Mock AIME 2024–2025 — Kimi K2.5 92.2%, GPT-5.4 mini 88.9%, Qwen3.5 397B-A17B 88.9%.

Which is better for coding?

There are no published SWE-bench Verified results for GPT-5.4 mini and Qwen3.5 397B-A17B yet, so there is no like-for-like coding score. On overall capability, GPT-5.4 mini 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 mini has the largest context window at 400,000 tokens, against 262,144 for Kimi K2.5 and 262,144 for Qwen3.5 397B-A17B. Maximum output per response: GPT-5.4 mini up to 128,000, Kimi K2.5 up to 262,144, Qwen3.5 397B-A17B up to 65,536 tokens.

Which can read images, PDFs, audio or video?

GPT-5.4 mini accepts text and images; Kimi K2.5 accepts text, images and video; Qwen3.5 397B-A17B accepts text, images and video. Kimi K2.5 handles the widest range of inputs.

Are any of these open source?

Kimi K2.5 and Qwen3.5 397B-A17B publishes its weights and can be self-hosted; GPT-5.4 mini is proprietary.

Which is newer?

GPT-5.4 mini is the newest, released Mar 17, 2026. Qwen3.5 397B-A17B came out Feb 15, 2026; Kimi K2.5 came out Jan 27, 2026. Knowledge cutoff: GPT-5.4 mini Aug 31, 2025, Kimi K2.5 Jan 2025.

How do you decide the winner?

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 18, 2026.