Mistral Medium 3 launches at $0.4/$2 per million tokens, matching 90% of Claude 3.7 Sonnet performance
Mistral AI launched Mistral Medium 3 on May 7, 2025, priced at $0.4 per million input tokens and $2 per million output tokens. The company claims the model performs at or above 90% of Claude Sonnet 3.7 on benchmarks while being significantly less expensive, and surpasses Llama 4 Maverick and Cohere Command A.
Mistral Medium 3 — Quick Specs
Mistral Medium 3 launches at $0.4/$2 per million tokens, matching 90% of Claude 3.7 Sonnet performance
Mistral AI launched Mistral Medium 3 on May 7, 2025, priced at $0.4 per million input tokens and $2 per million output tokens. According to Mistral, the model performs at or above 90% of Claude Sonnet 3.7 on benchmarks while costing 8x less.
Pricing and deployment
Mistral Medium 3 is available immediately on Mistral La Plateforme and Amazon Sagemaker, with IBM WatsonX, NVIDIA NIM, Azure AI Foundry, and Google Cloud Vertex support coming soon. The model can be self-hosted on environments with four GPUs or more.
The company claims the model beats DeepSeek v3 on pricing for both API and self-deployed systems, though specific comparative figures were not provided.
Performance claims
According to Mistral's internal benchmarks, Medium 3 surpasses:
- Llama 4 Maverick (open source)
- Cohere Command A (enterprise model)
- Claude Sonnet 3.7 at 90% of performance in most categories
Mistral states the model "comes close to its very large and much slower competitors" in coding and STEM tasks, and that third-party human evaluations show "much better performance" in coding compared to larger models. Specific benchmark scores were not disclosed in the announcement.
Enterprise features
Mistral Medium 3 supports:
- Hybrid, on-premises, or in-VPC deployment
- Custom post-training and continuous pretraining
- Full fine-tuning capabilities
- Integration with enterprise knowledge bases
Beta customers in financial services, energy, and healthcare are testing the model for customer service, business process personalization, and complex dataset analysis, according to Mistral.
Model architecture
Mistral did not disclose parameter count, context window size, training data cutoff date, or detailed architecture specifications for Medium 3.
What this means
Mistral Medium 3 represents aggressive pricing in the enterprise AI model market, undercutting competitors by significant margins if the performance claims hold. At $0.4/$2 per million tokens with claimed near-parity to Claude Sonnet 3.7, this pricing could pressure other providers to reduce costs or differentiate on capabilities beyond benchmarks.
The emphasis on self-hosting and enterprise customization positions Medium 3 for organizations requiring on-premises deployment or extensive fine-tuning—use cases where API-only models face adoption barriers. Mistral's teaser about a "large" model launch in coming weeks suggests a three-tier lineup (Small, Medium, Large) to compete across market segments.
The lack of disclosed benchmark scores, parameter count, and context window makes independent verification of claims impossible at launch. Performance relative to Claude 3.7 and Llama 4 Maverick will need external validation.
Related Articles
DeepSeek Releases Experimental V4-Flash-Vision-Exp, Claims Near-Parity With Opus 4.8 on Agent Benchmarks
DeepSeek has released V4-Flash-Vision-Exp, an experimental multimodal extension of V4-Flash that adds image understanding while preserving text reasoning capabilities. The company claims the model approaches or beats Anthropic's Opus 4.8 on its internal multimodal agent benchmarks.
Google Launches Gemini 3.5 Transcribe, a Speech-to-Text Model That Cleans Up Rambling Speech
Google has released Gemini 3.5 Transcribe, a new speech-to-text model that automatically detects over 85 languages, removes filler words, and structures unstructured speech into clean text. The model powers Android's Rambler feature and is rolling out to Chrome, Docs, Gmail, and other Google products.
Google DeepMind Launches Gemini 3.5 Transcribe, Claims 2.6% Word Error Rate in Testing
Google DeepMind has released Gemini 3.5 Transcribe, a speech-to-text model available via two APIs for real-time streaming and pre-recorded audio. According to Artificial Analysis benchmarks cited by Google, the model achieves a 2.6% word error rate for non-streaming transcription and 4.0% for streaming.
Google Launches Gemini 3.5 Transcribe with 2.6% Word Error Rate, Powers Gboard Rambler
Google has released Gemini 3.5 Transcribe, a speech-to-text model claiming a 4.0% word error rate in streaming mode and 2.6% in non-streaming mode, according to benchmarks from Artificial Analysis. The model already powers Gboard Rambler on Android and the Gemini app for macOS, with Chrome support coming next.
Comments
Loading...