ZMIME
Comparison · 3 models · Updated Nov 18, 2026

DeepSeek V3.2 vs GPT-5 Mini vs Kimi K2 Thinking

GPT-5 Mini comes out ahead, 70 to 61 and 60 on our weighted score, though DeepSeek V3.2 is 2.2× cheaper per token.

Price in rating: Excluded · Include price

  1. DeepSeek

    DeepSeek V3.2

    Released Dec 1, 2025

    60/100
    • ECI146.3
    • Price$0.28 / $0.42
    • Context128K
  2. Our pick

    OpenAI

    GPT-5 Mini

    Released Aug 7, 2025

    70/100
    • ECI146.0
    • Price$0.25 / $2.00
    • Context400K
  3. Moonshot AI

    Kimi K2 Thinking

    Released Nov 6, 2025

    61/100
    • ECI145.5
    • Price$0.60 / $2.50
    • Context262K
01 — Verdict

GPT-5 Mini is our pick

GPT-5 Mini is the better all-round choice, scoring 70/100 against Kimi K2 Thinking (61) and DeepSeek V3.2 (60). It leads on inputs & features and context window. The score weighs capability 67%, inputs & features 20%, context window 13%.

  • CapabilityDeepSeek V3.2Capabilities Index (ECI): DeepSeek V3.2 146.3 · GPT-5 Mini 146.0 · Kimi K2 Thinking 145.5
  • Lowest priceDeepSeek V3.2DeepSeek V3.2 $0.315 · GPT-5 Mini $0.688 · Kimi K2 Thinking $1.07 per 1M tokens (3:1 blend)
  • Longest contextGPT-5 MiniGPT-5 Mini 400,000 · Kimi K2 Thinking 262,144 · DeepSeek V3.2 128,000 tokens
  • Widest inputsGPT-5 MiniDeepSeek V3.2: Text · GPT-5 Mini: Text, Images · Kimi K2 Thinking: Text
  • Self-hostingDeepSeek V3.2 and Kimi K2 ThinkingPublishes downloadable weights (MIT License)
How the score is built
MeasureWeightDeepSeek V3.2GPT-5 MiniKimi K2 Thinking
CapabilityCapabilities Index (ECI)67%747473
Inputs & features20%357035
Context window13%325043
Overall100%60/10070/10061/100
02 — Side by side

Every spec in one table

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

DeepSeek V3.2 vs GPT-5 Mini vs Kimi K2 Thinking specifications side by side
SpecificationDeepSeek V3.2DeepSeekGPT-5 MiniOpenAIKimi K2 ThinkingMoonshot AI
Capability
Capabilities Index (ECI)146.3 (best)146.0145.5
ECI rank#87 of 158 (best)#88 of 158#90 of 158
GPQA DiamondGraduate-level science questions83.4%75.0%84.2% (best)
FrontierMath Tiers 1–3Research-level mathematics—46.7%—
OTIS Mock AIME 2024–2025Competition mathematics87.8% (best)86.7%83.1%
SWE-bench VerifiedFixing real GitHub issues—64.7%—
SimpleQA VerifiedShort factual questions—21.6%—
Price per million tokens
Input$0.28$0.25 (best)$0.60
Output$0.42 (best)$2.00$2.50
Cached input—$0.025—
Blended (3:1)$0.315 (best)$0.688$1.07
Long-context rateSame rateSame rateSame rate
Price sourceMedian of 30 providersOfficial OpenAI APIMedian of 14 providers
Limits
Context window128,000 tokens400,000 tokens (best)262,144 tokens
Max output64,000 tokens128,000 tokens262,144 tokens (best)
Inputs and features
TextYesYesYes
ImagesNoYesNo
PDFsNoNoNo
AudioNoNoNo
VideoNoNoNo
ReasoningYesYesminimal · low · medium · highYes
Tool callingYesYesYes
Structured outputNoYesNo
Availability
WeightsOpenMIT LicenseProprietaryOpen
API model ID—gpt-5-mini—
API providers3036 (best)15
ReleasedDec 1, 2025Aug 7, 2025Nov 6, 2025
Knowledge cutoffJul 2024May 30, 2024Aug 2024
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.

  • DeepSeek V3.2$3.64
  • GPT-5 Mini$6.50
  • Kimi K2 Thinking$11.00
04 — Questions

Which should you choose?

Which is better: DeepSeek V3.2, GPT-5 Mini or Kimi K2 Thinking?

GPT-5 Mini is the better all-round choice, scoring 70/100 against Kimi K2 Thinking (61) and DeepSeek V3.2 (60). It leads on inputs & features and context window. The score weighs capability 67%, inputs & features 20%, context window 13%.

Which is cheaper, DeepSeek V3.2, GPT-5 Mini or Kimi K2 Thinking?

DeepSeek V3.2 is cheaper at $0.28 input / $0.42 output per million tokens (median across 30 API providers). GPT-5 Mini costs $0.25 input / $2.00 output per million tokens (official OpenAI API price); Kimi K2 Thinking costs $0.60 input / $2.50 output per million tokens (median across 14 API providers). At a typical mix of three input tokens to one output token, that is $0.315 per million tokens for DeepSeek V3.2 versus $0.688 for GPT-5 Mini (2.2× as much) and $1.07 for Kimi K2 Thinking (3.4× as much).

Which scores higher on benchmarks?

DeepSeek V3.2 scores higher on the Capabilities Index (ECI): DeepSeek V3.2 146.3 (#87 of 158), GPT-5 Mini 146.0 (#88 of 158) and Kimi K2 Thinking 145.5 (#90 of 158). The confidence ranges of the top two overlap (144.8–147.8 vs 144.4–147.6), so treat the gap as small. On individual benchmarks: GPQA Diamond — Kimi K2 Thinking 84.2%, DeepSeek V3.2 83.4%, GPT-5 Mini 75.0%; OTIS Mock AIME 2024–2025 — DeepSeek V3.2 87.8%, GPT-5 Mini 86.7%, Kimi K2 Thinking 83.1%.

Which is better for coding?

There are no published SWE-bench Verified results for DeepSeek V3.2 and Kimi K2 Thinking yet, so there is no like-for-like coding score. On overall capability, DeepSeek V3.2 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 Mini has the largest context window at 400,000 tokens, against 262,144 for Kimi K2 Thinking and 128,000 for DeepSeek V3.2. Maximum output per response: DeepSeek V3.2 up to 64,000, GPT-5 Mini up to 128,000, Kimi K2 Thinking up to 262,144 tokens.

Which can read images, PDFs, audio or video?

DeepSeek V3.2 accepts text; GPT-5 Mini accepts text and images; Kimi K2 Thinking accepts text. GPT-5 Mini handles the widest range of inputs.

Are any of these open source?

DeepSeek V3.2 and Kimi K2 Thinking publishes its weights (MIT License) and can be self-hosted; GPT-5 Mini is proprietary.

Which is newer?

DeepSeek V3.2 is the newest, released Dec 1, 2025. Kimi K2 Thinking came out Nov 6, 2025; GPT-5 Mini came out Aug 7, 2025. Knowledge cutoff: DeepSeek V3.2 Jul 2024, GPT-5 Mini May 30, 2024, Kimi K2 Thinking Aug 2024.

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.