ZMIME
Comparison · 3 models · Updated Nov 22, 2026

Mistral Large vs Llama-3.2-90B-Vision-Instruct vs Gemma 2 27B

Too close to call on our weighted score (Llama-3.2-90B-Vision-Instruct 49, Mistral Large 46, Gemma 2 27B 40). The right pick depends on what you value most.

  1. Mistral AI

    Mistral Large

    Released Feb 26, 2024

    46/100
    • ECI120.0
    • Price$2.00 / $6.00
    • Context128K
  2. Meta

    Llama-3.2-90B-Vision-Instruct

    Released Sep 25, 2024

    49/100
    • ECI122.9
    • Price—
    • Context128K
  3. Google

    Gemma 2 27B

    Released Jul 13, 2024

    40/100
    • ECI118.6
    • Price$0.65 / $0.65
    • Context8K
  4. Mistral AI

    Mistral Large

    Released Feb 26, 2024

    46/100
    • ECI120.0
    • Price$2.00 / $6.00
    • Context128K
  5. Meta

    Llama-3.2-90B-Vision-Instruct

    Released Sep 25, 2024

    49/100
    • ECI122.9
    • Price—
    • Context128K
  6. Google

    Gemma 2 27B

    Released Jul 13, 2024

    40/100
    • ECI118.6
    • Price$0.65 / $0.65
    • Context8K
01 — Verdict

Too close to call

It is close. Our weighted score puts them within 2 points (Llama-3.2-90B-Vision-Instruct 49/100, Mistral Large 46/100, Gemma 2 27B 40/100), so choose by what matters most for your work: Llama-3.2-90B-Vision-Instruct for raw capability and Gemma 2 27B on price. The score weighs capability 83%, context window 17%.

  • CapabilityLlama-3.2-90B-Vision-InstructCapabilities Index (ECI): Llama-3.2-90B-Vision-Instruct 122.9 · Mistral Large 120.0 · Gemma 2 27B 118.6
  • Lowest priceGemma 2 27BGemma 2 27B $0.65 · Mistral Large $3.00 per 1M tokens (3:1 blend) · Llama-3.2-90B-Vision-Instruct unpriced
  • Longest contextMistral Large and Llama-3.2-90B-Vision-InstructMistral Large 128,000 · Llama-3.2-90B-Vision-Instruct 128,000 · Gemma 2 27B 8,192 tokens
  • Widest inputsMistral Large and Llama-3.2-90B-Vision-InstructMistral Large: Text, PDFs · Llama-3.2-90B-Vision-Instruct: Text, Images · Gemma 2 27B: Text
  • Self-hostingLlama-3.2-90B-Vision-Instruct and Gemma 2 27BPublishes downloadable weights (Llama 3.2 Community License)
How the score is built
MeasureWeightMistral LargeLlama-3.2-90B-Vision-InstructGemma 2 27B
CapabilityCapabilities Index (ECI)83%495248
Context window17%32320
Overall100%46/10049/10040/100

Left out because at least one model lacks the data: price and inputs & features. The remaining weights were rescaled.

How the score is built
MeasureWeightMistral LargeLlama-3.2-90B-Vision-InstructGemma 2 27B
CapabilityCapabilities Index (ECI)83%495248
Context window17%32320
Overall100%46/10049/10040/100

Left out because at least one model lacks the data: inputs & features. The remaining weights were rescaled.

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.

Mistral Large vs Llama-3.2-90B-Vision-Instruct vs Gemma 2 27B specifications side by side
SpecificationMistral LargeMistral AILlama-3.2-90B-Vision-InstructMetaGemma 2 27BGoogle
Capability
Capabilities Index (ECI)120.0122.9 (best)118.6
ECI rank#151 of 159#149 of 159 (best)#153 of 159
ECI confidence interval115.9–123.6119.2–126.3114.2–122.4
ECI evaluation dateFeb 26, 2024Sep 24, 2024Jun 24, 2024
Documented capabilitiesFlagship Mistral model for advanced reasoning, coding, and multilingual workLarge open multimodal Llama model for image reasoning, document QA, and captioningOpen Gemma instruction model for efficient chat and self-hosted deployments
Model parametersNot publishedNot publishedNot published
Output modalitiesTextTextText
GPQA DiamondGraduate-level science questions38.8%41.0% (best)36.5%
OTIS Mock AIME 2024–2025Competition mathematics1.9%2.6% (best)1.4%
Price per million tokens
Input$2.00Not published$0.65 (best)
Output$6.00Not published$0.65 (best)
Cached inputNot publishedNot publishedNot published
Blended (3:1)$3.00Not published$0.65 (best)
Long-context rateSame rateNot publishedSame rate
Price sourceMedian of 8 providersNot publishedMedian of 4 providers
Limits
Context window128,000 tokens (best)128,000 tokens (best)8,192 tokens
Max inputNot publishedNot publishedNot published
Max output102,400 tokens (best)4,096 tokens2,048 tokens
Inputs and features
TextYesYesYes
ImagesNoYesNo
PDFsYesNoNo
AudioNoNoNo
VideoNoNoNo
ReasoningNoNoNo
Tool callingYesYesNo
Structured outputNot publishedNot publishedNot published
File attachmentsYesYesNo
Temperature controlYesYesYes
Reasoning controlNot publishedNot publishedNot published
Availability
WeightsProprietaryOpenLlama 3.2 Community LicenseOpen
API model IDNot publishedNot publishedNot published
API providers8 (best)14
ReleasedFeb 26, 2024Sep 25, 2024Jul 13, 2024
Knowledge cutoffNov 30, 2024Dec 2023Jun 30, 2024
Publisher documentationNot publishedModel weightsModel weights
Source metadata updatedFeb 26, 2024Sep 25, 2024Jul 13, 2024
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.

  • Mistral Large$32.03
  • Llama-3.2-90B-Vision-Instruct—
  • Gemma 2 27B$7.80
04 — Questions

Which should you choose?

Which is better: Mistral Large, Llama-3.2-90B-Vision-Instruct or Gemma 2 27B?

It is close. Our weighted score puts them within 2 points (Llama-3.2-90B-Vision-Instruct 49/100, Mistral Large 46/100, Gemma 2 27B 40/100), so choose by what matters most for your work: Llama-3.2-90B-Vision-Instruct for raw capability and Gemma 2 27B on price. The score weighs capability 83%, context window 17%.

Which is cheaper, Mistral Large, Llama-3.2-90B-Vision-Instruct or Gemma 2 27B?

Gemma 2 27B is cheaper at $0.65 input / $0.65 output per million tokens (median across 4 API providers). Mistral Large costs $2.00 input / $6.00 output per million tokens (median across 8 API providers). At a typical mix of three input tokens to one output token, that is $0.65 per million tokens for Gemma 2 27B versus $3.00 for Mistral Large (4.6× as much). Llama-3.2-90B-Vision-Instruct has no published per-token price.

Which scores higher on benchmarks?

Llama-3.2-90B-Vision-Instruct scores higher on the Capabilities Index (ECI): Llama-3.2-90B-Vision-Instruct 122.9 (#149 of 159), Mistral Large 120.0 (#151 of 159) and Gemma 2 27B 118.6 (#153 of 159). The confidence ranges of the top two overlap (119.2–126.3 vs 115.9–123.6), so treat the gap as small. On individual benchmarks: GPQA Diamond — Llama-3.2-90B-Vision-Instruct 41.0%, Mistral Large 38.8%, Gemma 2 27B 36.5%; OTIS Mock AIME 2024–2025 — Llama-3.2-90B-Vision-Instruct 2.6%, Mistral Large 1.9%, Gemma 2 27B 1.4%.

Which is better for coding?

There are no published SWE-bench Verified results for Mistral Large, Llama-3.2-90B-Vision-Instruct and Gemma 2 27B yet, so there is no like-for-like coding score. On overall capability, Llama-3.2-90B-Vision-Instruct leads, which tends to carry over to coding, but test on your own codebase. Note that Gemma 2 27B has no confirmed tool-calling support in this comparison.

Which has the bigger context window?

Mistral Large and Llama-3.2-90B-Vision-Instruct have the largest context windows (128,000 and 128,000 tokens), against 8,192 for Gemma 2 27B. Maximum output per response: Mistral Large up to 102,400, Llama-3.2-90B-Vision-Instruct up to 4,096, Gemma 2 27B up to 2,048 tokens.

Which can read images, PDFs, audio or video?

Mistral Large accepts text and PDFs; Llama-3.2-90B-Vision-Instruct accepts text and images; Gemma 2 27B accepts text. Mistral Large handles the widest range of inputs.

Are any of these open source?

Llama-3.2-90B-Vision-Instruct and Gemma 2 27B publishes its weights (Llama 3.2 Community License) and can be self-hosted; Mistral Large is proprietary.

Which is newer?

Llama-3.2-90B-Vision-Instruct is the newest, released Sep 25, 2024. Gemma 2 27B came out Jul 13, 2024; Mistral Large came out Feb 26, 2024. Knowledge cutoff: Mistral Large Nov 30, 2024, Llama-3.2-90B-Vision-Instruct Dec 2023, Gemma 2 27B Jun 30, 2024.

How do you decide the winner?

This rating uses capability (83%, independent benchmark results); context window (17%, 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.