ZMIME
Comparison · 3 models · Updated Nov 18, 2026

Gemma 2 27B vs Claude Haiku 3 vs Kimi K2 Instruct

Kimi K2 Instruct comes out ahead, 55 to 48 and 32 on our weighted score, though Claude Haiku 3 is 2× cheaper per token.

Price in rating: Excluded · Include price

  1. Google

    Gemma 2 27B

    Released Jul 13, 2024

    32/100
    • ECI118.6
    • Price$0.65 / $0.65
    • Context8K
  2. Anthropic

    Claude Haiku 3

    Released Mar 13, 2024

    48/100
    • ECI117.8
    • Price$0.25 / $1.25
    • Context200K
  3. Our pick

    Moonshot AI

    Kimi K2 Instruct

    Released Jul 11, 2025

    55/100
    • ECI141.3
    • Price$0.57 / $2.30
    • Context131K
01 — Verdict

Kimi K2 Instruct is our pick

Kimi K2 Instruct is the better all-round choice, scoring 55/100 against Claude Haiku 3 (48) and Gemma 2 27B (32). It leads on capability. Claude Haiku 3 wins on inputs & features and context window. The score weighs capability 67%, inputs & features 20%, context window 13%.

  • CapabilityKimi K2 InstructCapabilities Index (ECI): Kimi K2 Instruct 141.3 · Gemma 2 27B 118.6 · Claude Haiku 3 117.8
  • Lowest priceClaude Haiku 3Claude Haiku 3 $0.50 · Gemma 2 27B $0.65 · Kimi K2 Instruct $1.00 per 1M tokens (3:1 blend)
  • Longest contextClaude Haiku 3Claude Haiku 3 200,000 · Kimi K2 Instruct 131,072 · Gemma 2 27B 8,192 tokens
  • Widest inputsClaude Haiku 3Gemma 2 27B: Text · Claude Haiku 3: Text, Images, PDFs · Kimi K2 Instruct: Text
  • Self-hostingGemma 2 27B and Kimi K2 InstructPublishes downloadable weights (Modified MIT)
How the score is built
MeasureWeightGemma 2 27BClaude Haiku 3Kimi K2 Instruct
CapabilityCapabilities Index (ECI)67%484769
Inputs & features20%06025
Context window13%03932
Overall100%32/10048/10055/100
02 — Side by side

Every spec in one table

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

Gemma 2 27B vs Claude Haiku 3 vs Kimi K2 Instruct specifications side by side
SpecificationGemma 2 27BGoogleClaude Haiku 3AnthropicKimi K2 InstructMoonshot AI
Capability
Capabilities Index (ECI)118.6117.8141.3 (best)
ECI rank#152 of 158#154 of 158#108 of 158 (best)
GPQA DiamondGraduate-level science questions36.5% (best)36.3%—
OTIS Mock AIME 2024–2025Competition mathematics1.4%1.8% (best)—
Price per million tokens
Input$0.65$0.25 (best)$0.57
Output$0.65 (best)$1.25$2.30
Cached input———
Blended (3:1)$0.65$0.50 (best)$1.00
Long-context rateSame rateSame rateSame rate
Price sourceMedian of 3 providersMedian of 3 providersMedian of 11 providers
Limits
Context window8,192 tokens200,000 tokens (best)131,072 tokens
Max output2,048 tokens4,096 tokens32,768 tokens (best)
Inputs and features
TextYesYesYes
ImagesNoYesNo
PDFsNoYesNo
AudioNoNoNo
VideoNoNoNo
ReasoningNoNoNo
Tool callingNoYesYes
Structured outputNoNoNo
Availability
WeightsOpenProprietaryOpenModified MIT
API model ID———
API providers3311 (best)
ReleasedJul 13, 2024Mar 13, 2024Jul 11, 2025
Knowledge cutoffJun 30, 2024Aug 31, 2023—
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.

  • Gemma 2 27B$7.80
  • Claude Haiku 3$5.00
  • Kimi K2 Instruct$10.30
04 — Questions

Which should you choose?

Which is better: Gemma 2 27B, Claude Haiku 3 or Kimi K2 Instruct?

Kimi K2 Instruct is the better all-round choice, scoring 55/100 against Claude Haiku 3 (48) and Gemma 2 27B (32). It leads on capability. Claude Haiku 3 wins on inputs & features and context window. The score weighs capability 67%, inputs & features 20%, context window 13%.

Which is cheaper, Gemma 2 27B, Claude Haiku 3 or Kimi K2 Instruct?

Claude Haiku 3 is cheaper at $0.25 input / $1.25 output per million tokens (median across 3 API providers). Gemma 2 27B costs $0.65 input / $0.65 output per million tokens (median across 3 API providers); Kimi K2 Instruct costs $0.57 input / $2.30 output per million tokens (median across 11 API providers). At a typical mix of three input tokens to one output token, that is $0.50 per million tokens for Claude Haiku 3 versus $0.65 for Gemma 2 27B (1.3× as much) and $1.00 for Kimi K2 Instruct (2× as much).

Which scores higher on benchmarks?

Kimi K2 Instruct scores higher on the Capabilities Index (ECI): Kimi K2 Instruct 141.3 (#108 of 158), Gemma 2 27B 118.6 (#152 of 158) and Claude Haiku 3 117.8 (#154 of 158). Their confidence ranges do not overlap (138.3–144.0 vs 114.5–122.4), so the gap is a real one.

Which is better for coding?

There are no published SWE-bench Verified results for Gemma 2 27B, Claude Haiku 3 and Kimi K2 Instruct yet, so there is no like-for-like coding score. On overall capability, Kimi K2 Instruct leads, which tends to carry over to coding, but test on your own codebase. Note that Gemma 2 27B does not support tool calling, which most coding agents need.

Which has the bigger context window?

Claude Haiku 3 has the largest context window at 200,000 tokens, against 131,072 for Kimi K2 Instruct and 8,192 for Gemma 2 27B. Maximum output per response: Gemma 2 27B up to 2,048, Claude Haiku 3 up to 4,096, Kimi K2 Instruct up to 32,768 tokens.

Which can read images, PDFs, audio or video?

Gemma 2 27B accepts text; Claude Haiku 3 accepts text, images and PDFs; Kimi K2 Instruct accepts text. Claude Haiku 3 handles the widest range of inputs.

Are any of these open source?

Gemma 2 27B and Kimi K2 Instruct publishes its weights (Modified MIT) and can be self-hosted; Claude Haiku 3 is proprietary.

Which is newer?

Kimi K2 Instruct is the newest, released Jul 11, 2025. Gemma 2 27B came out Jul 13, 2024; Claude Haiku 3 came out Mar 13, 2024. Knowledge cutoff: Gemma 2 27B Jun 30, 2024, Claude Haiku 3 Aug 31, 2023.

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.