AI Models · Open-Source vs Paid

Mistral Medium 3.5 Open vs Mistral Large 3 2512 (batch) Paid

Mistral Medium 3.5 vs Mistral Large 3 2512 (batch) compared — price per token, context window, multimodality, openness and which to choose. Can the open-source model replace the paid one? Full 2026 breakdown.

Mistral Large 3 2512 (batch) — full profile ›

Prices & specs refreshed from live data · olud.ai

Open-model prices = cheapest provider via OpenRouter; official maker rates may be higher.

Mistral Medium 3.5OpenMistral AI
$1.5 /M input
$7.5 /M outputNo per-token fees if you self-host
TypeOpen-weight
Context window262K tokens
MultimodalYes
Self-hostYes
Mistral Large 3 2512 (batch)PaidMistral AI
$0.25 /M input
$0.75 /M outputManaged API (no infra to run)
TypeProprietary
Context window262K tokens
MultimodalYes
Self-hostNo
Choose Mistral Medium 3.5 if you want to self-host, keep your data private and skip per-token fees — it's open-weight and runs on your own hardware. Choose Mistral Large 3 2512 (batch) if you want frontier capability through a managed API with zero infrastructure to run.

Mistral Medium 3.5 vs Mistral Large 3 2512 (batch) specs

SpecMistral Medium 3.5Mistral Large 3 2512 (batch)Winner
MakerMistral AIMistral AI–
TypeOpen-weightProprietaryMistral Medium 3.5
Context window262K tokens262K tokens= Tie
Input price$1.5/M · free self-host$0.25/MMistral Large 3 2512 (batch)
Output price$7.5/M · free self-host$0.75/MMistral Large 3 2512 (batch)
Vision / multimodalYesYes= Tie
Tool / function callingYesYes= Tie
Self-hostableYesNo (API only)Mistral Medium 3.5
LicenseApache 2.0ProprietaryMistral Medium 3.5

Price gap & when to choose each

10×cheaper per output token

Mistral Large 3 2512 (batch) is ~10× cheaper than Mistral Medium 3.5 on output tokens ($0.75 vs $7.5 per M tokens).

Choose Mistral Medium 3.5 if…
  • You want to self-host or run on your cloud
  • You prioritize data privacy & control
  • You are building open or reproducible AI
Choose Mistral Large 3 2512 (batch) if…
  • You want frontier performance through a managed API
  • You want the lower output price
  • You value reliability & ecosystem
  • You don't want to manage infrastructure

Feature comparison

CapabilityMistral Medium 3.5Mistral Large 3 2512 (batch)
Open weights (downloadable)✓✗
Self-hostable✓✗
Runs fully offline✓✗
Vision / multimodal✓✓
Tool / function calling✓✓
1M+ context window✗✗

Benchmarks: Mistral Medium 3.5 vs Mistral Large 3 2512 (batch)

Independent benchmark scores measured by Artificial Analysis. Higher is better (except latency).

Mistral Large 3 2512 (batch) delivers 2.6× more intelligence per dollar.
Mistral Medium 3.5
Mistral Large 3 2512 (batch)
Intelligence index
14.9
9.7
Coding index
46.9
20.1
GPQA
74.8%
68%
Humanity's Last Exam
13.8%
4.2%
Long Context Reasoning
69.3%
36%
SciCode
40.2%
36.6%
IFBench
68.8%
36.2%
τ²-Bench
94.2%
24.6%
τ-Bench Banking
15.1%
5.8%
Terminal-Bench
50.6%
12%
Terminal-Bench Hard
33.3%
15.9%
Math index
–
38
MMLU-Pro
–
80.7%
LiveCodeBench
–
46.5%
AIME 2025
–
38%
Speed
142.2 tok/s
81.3 tok/s
Latency
0.74s
0.68s
Intelligence per $
5
12.9

Benchmark data by Artificial Analysis.

How Mistral Medium 3.5 and Mistral Large 3 2512 (batch) score

🏆 Best value & openness: Mistral Medium 3.5 (4.6 vs 3.3 / 5)
CriterionMistral Medium 3.5Mistral Large 3 2512 (batch)
Cost-efficiency4.05.0
Context window4.04.0
Openness5.01.5
Self-hosting5.01.0
Multimodality5.05.0

Scores come from live data — output price (cost), context length, open vs closed weights (openness & self-hosting) and vision/tool support (multimodality). Raw task quality isn't scored here; it depends on your benchmark — see the verdict.

What each model is

Mistral Medium 3.5 Open

Mistral AI · Open-weight

Mistral Medium 3.5 is a dense 128B instruction-following model from Mistral AI. It supports text and image inputs with text output, and is designed for agentic workflows, coding, and complex...

Mistral Large 3 2512 (batch) Paid

Mistral AI · Proprietary

Mistral Large 3 2512 is Mistral’s most capable model to date, featuring a sparse mixture-of-experts architecture with 41B active parameters (675B total), and released under the Apache 2.0 license.

Other models in these families

These variants are tracked but not compared here — one page per family keeps the comparison readable.

Other variants tracked
Mistral Medium 3.5 (batch)Mistral Small 4Mistral Small 4 (batch)Mistral Medium 3.1Mistral Medium 3.1 (batch)Mistral Large 3 2512Mistral Large 3 2512 (batch)Devstral 2 2512Mistral Medium 3Mistral Small 3.1 24BMistral Small 3.2 24BMistral LargeMistral Large 2407Mistral Small 3SabaMinistral 3 14B 2512Mixtral 8x22B InstructMinistral 3 8B 2512Ministral 3 8B 2512 (batch)Ministral 3 3B 2512UncensoredCodestral 2508Codestral 2508 (batch)Voxtral Small 24B 2507Mistral Nemo

Frequently asked questions

Is Mistral Medium 3.5 as good as Mistral Large 3 2512 (batch)?

Mistral Medium 3.5 is open-weight and competitive on many tasks, but Mistral Large 3 2512 (batch) may still lead on the hardest reasoning and agentic work. The gap keeps narrowing — benchmark both on your actual use case before deciding.

Can I run Mistral Medium 3.5 locally?

Yes. Mistral Medium 3.5 has open weights, so you can self-host it on your own GPUs or run it via a low-cost API. Mistral Large 3 2512 (batch) is API-only and cannot be self-hosted.

Mistral Medium 3.5 vs Mistral Large 3 2512 (batch) — which should I pick in 2026?

Choose Mistral Medium 3.5 if you want to self-host, keep your data private and skip per-token fees — it's open-weight and runs on your own hardware. Choose Mistral Large 3 2512 (batch) if you want frontier capability through a managed API with zero infrastructure to run.

People also compare

Hy3 vs Mistral Large 3 2512 (batch)GLM 5.3 vs Mistral Large 3 2512 (batch)DeepSeek V4.1 Flash vs Mistral Large 3 2512 (batch)MiMo-V2.5-Pro vs Mistral Large 3 2512 (batch)Nemotron 3 Ultra vs Mistral Large 3 2512 (batch)Kimi K3 (batch) vs Mistral Large 3 2512 (batch)

Explore more open-source AI

Browse the full open-source model leaderboard, thousands of tools and live benchmarks — all in one place.

Open the leaderboard →