Mistral Large 4 vs Claude, GPT, Gemini and DeepSeek: Pricing

Update (9 Nov 2026): Claude Fable 5.2 and Mythos 5.2 are out - prices, access and benchmarks are in my Claude Fable 5.2 guide.

👁 124 reads

Mistral Large 4 costs $1.36 per million input tokens and $4.18 per million output tokens - less than Claude Sonnet 5.5, GPT-6.1 Sol and Gemini 3.1 Pro, but significantly more than DeepSeek V4. In the independent Artificial Analysis Intelligence Index it scores 38.4 points versus 57.6 for Claude Opus 5.5, and in our test (conducted in Polish) it scored 69.2 points versus 80.4 for Sonnet 5.5 and 74.1 for DeepSeek V4 Pro. We’ve gathered pricing and results in one place. Full launch details are in our news article Mistral Large 4.

API Pricing: Cost Per Million Tokens

Gemini 4 Argon API pricing is set to launch at $2 / $10 USD (eventually $4 / $20 USD) - we cover this in our article Gemini 4 Argon. It’s not yet on OpenRouter, so it’s not in our test. DeepSeek pricing changes frequently, sometimes several times a week.

Key takeaways:

• Versus Claude and GPT: Large 4 is roughly 1.5x cheaper on input and 2.4x cheaper on output than Sonnet 5.5 and GPT-6.1 Sol, and with the promotional pricing 3x and nearly 5x cheaper respectively.
• Versus DeepSeek: DeepSeek V4 Pro on OpenRouter is currently 6.5x cheaper on input and 10x cheaper on output than Large 4’s list price.
• Versus predecessor: Medium 3.5 was more expensive on output ($7.50 USD), so Large 4 is a cheaper and more powerful Mistral flagship.
• Watch out for reasoning: you also pay for “thinking” tokens. In our test, Large 4 with reasoning cost $0.134 for 13 tasks, while without reasoning it cost $0.015 - almost 9x less.

Artificial Analysis: Independent Index

Data from October 6 via Trending Topics. Artificial Analysis currently counts Large 4 as a closed model because the weights aren’t available yet. After their release, it would rank 8th among open models - behind seven Chinese models and ahead of DeepSeek V4 Pro. This is the highest score for an open-weight model from Europe or the USA, which aligns with Mistral’s claims.

Our Polish-Language Test: Same 13 Tasks

We described the methodology, all partial scores and error examples in our article Mistral Large 4 po polsku. You can read all model responses side by side in our Arena modeli AI po polsku.

Value for Money

If you care about Polish text quality per dollar, Large 4 doesn’t perform well:

• Gemini 3.8 Flash scores 10.2 points higher at a lower list price ($0.75 / $3.75 USD).
• DeepSeek V4 Pro scores 4.9 points higher at a price several times lower.
• Sonnet 5.5 and GPT-6.1 Sol are 11-17 points better, but 2.4x more expensive on output.

Large 4’s advantages lie elsewhere: European provider, EU data processing, image input, promised open weights, and cybersecurity results (82% in AA Cyber Index for vulnerability reproduction, 93% in Cybench according to Mistral). For these use cases, none of the compared closed models will give you weights to run on your own infrastructure.

Which to Choose

This is our assessment as of October 7. Mistral has announced that it will publish a final checkpoint along with the weights, so we’ll repeat the test after October 27. A full overview of Mistral’s offerings is in our Mistral AI hub, and pricing for its chat and API is in our article Mistral cena.

Sources

• Mistral: Introducing Mistral Large 4 (October 6, 2026) and model documentation
• Trending Topics: Mistral Large 4 in Artificial Analysis (October 6, 2026)
• OpenRouter model list - pricing from October 7, 2026
• Quantslant Lab - test conducted October 7, 2026

Verified: October 7, 2026.

// read moreMistral Large 4 Enters the AI Arena: October 2026 RankingsBenchmarksMistral Large 4 (Le Chonk): Specs, Pricing, and Open WeightsNewsMistral AI Pricing: Vibe Pro & API Costs in 2026Guides

Kinetyra