Mistral's New Reasoning Model Magistral Falls Behind DeepSeek R1 Despite Speed Advantages

BigGo Editorial Team
Mistral's New Reasoning Model Magistral Falls Behind DeepSeek R1 Despite Speed Advantages

Mistral AI has released Magistral, their first reasoning model designed to compete with established players like OpenAI and DeepSeek. However, early community analysis reveals significant performance gaps that highlight the challenges facing European AI companies in the global race.

The French AI company launched Magistral in two versions: a 24 billion parameter open-source Small variant and a more powerful Medium enterprise version. While Mistral positions this as a breakthrough in transparent, multilingual reasoning, the community response has been mixed at best.

Performance Concerns Overshadow Launch

Benchmark comparisons show Magistral struggling against current state-of-the-art models. The community quickly noted that Magistral loses to DeepSeek R1 in every direct comparison, despite DeepSeek's model being significantly larger at 671 billion parameters with 37 billion active. This size difference makes the comparison somewhat unfair, but it underscores the resource constraints facing smaller AI companies.

The situation becomes more concerning when considering that Mistral likely tested against the older version of DeepSeek R1, not the newer release. At more than double the cost of competitors, Magistral's value proposition becomes questionable for many users.

Speed as the Hidden Advantage

Despite performance shortcomings, Magistral excels in one crucial area that Mistral buried deep in their announcement: speed. Users report response times of around one second compared to 5-8 seconds for competing models. This 10x speed improvement through their partnership with Cerebras creates a notably better user experience.

For me this is more important than quality. I love fast responses, feels more futuristic.

However, this speed advantage only applies to queries without web search, where Mistral becomes significantly slower, reducing the overall benefit for research-heavy tasks.

Magistral: A new reasoning model by Mistral AI, boasting improved speed in response times
Magistral: A new reasoning model by Mistral AI, boasting improved speed in response times

European AI's Uphill Battle

The lukewarm reception of Magistral reflects broader challenges facing European AI companies. With only 200 employees, Mistral competes against giants like OpenAI and Google, while even Anthropic boasts over 1,000 staff members. The funding landscape also differs dramatically, with European venture capital lacking the deep pockets of Silicon Valley investors.

Community discussions reveal frustration with Europe's position in the AI race. While the continent produces influential researchers and maintains strong enterprise software companies like SAP, consumer-facing AI platforms remain dominated by American firms. The regulatory environment, often cited as a hindrance, appears less significant than the fundamental resource and market access disadvantages.

Technical Implementation and Availability

Magistral incorporates several interesting technical choices, including removing KL divergence penalties during training to allow more aggressive learning. The model supports native reasoning across multiple languages including English, French, Spanish, German, Italian, Arabic, Russian, and Simplified Chinese.

The open-source Small version is available under Apache 2.0 license through Hugging Face, while the Medium variant requires enterprise licensing. Community members have already created optimized versions for consumer hardware, making local deployment possible on high-end gaming systems or 32GB MacBooks.

Looking Forward

Mistral's first reasoning model represents a solid engineering effort hampered by resource constraints and timing. While Magistral may not lead in raw performance, its speed advantages and multilingual capabilities could carve out specific use cases. The company promises rapid iteration and improvements, suggesting this release serves as a foundation rather than a final product.

For European AI ambitions, Magistral highlights both the potential and limitations of competing against well-funded American and Chinese rivals. Success may require focusing on specific strengths like speed and efficiency rather than attempting to match the scale of larger competitors.

Reference: Magistral