Mistral's Open-Source Model Le Chonk Claims to Surpass DeepSeek and Other Chinese Rivals
nashnova research
French AI firm Mistral released Le Chonk, a trillion-parameter open-source model it calls the world's best in automated coding and cybersecurity — free for all users — but security experts warn open-source cuts both ways.
What is Le Chonk, and what backs the "world's best" claim?
Le Chonk is Mistral's trillion-parameter open-source model, now free for all users.
Mistral says it tops DeepSeek and Z.ai on automated coding and cybersecurity benchmarks — claiming the global lead, not just "best outside China."
The training method: feeding the model progressively harder cybersecurity and coding problems until it caught and passed Chinese rivals. This means → Mistral didn't just scale parameters; it used targeted high-difficulty data to sharpen specific capabilities.
Why is Mistral making "beating Chinese models" the headline?
Mistral's VP of Science Pierre Stock called this "a pivotal moment in the global AI leadership race," aimed squarely at enterprise and government buyers.
He compared open-source to electrical infrastructure: "AI will be like electricity — you don't want to live in a world where your neighbour can cut your power." In plain terms = once model weights are published online, anyone can download them and no government can unilaterally cut access — that is the strategic value Mistral sees in open-source.
Stock also flagged Chinese models directly: "You cannot guarantee they won't exhibit malicious behaviour; they could become threat vectors." This reflects a broader move by European AI firms to frame "non-China alternative" as a core competitive narrative — courting both users and government contracts.
What do security experts say? Does open-source actually solve the problem?
Rafe Pilling, senior director of threat intelligence at security firm Sophos, pushed back: all advanced models can go wrong without strict oversight, regardless of nationality.
"The technical reality and the often nationalistic narrative are two different things," he told *Wired*. "What matters is building the right monitoring and governance mechanisms."
Pilling also highlighted the dual nature of open-source: "Any model that can be freely downloaded and shared can be misused just as easily as it can be used legitimately." This means → Le Chonk reduces Western dependence on Chinese models, but it also hands potential attackers a freely customisable tool — the security problem doesn't vanish just because the label says "open-source."
Open-source models — cybersecurity shield for the West, or a new attack surface?
BULL
Breaks dependency
Once weights are public, no government can unilaterally cut access.
Transparent and auditable
Open-source lets the security community inspect code and weights — hidden backdoors are harder to conceal.
BEAR
Attackers benefit too
Free download means malicious actors can customise attack tools just as easily.
Nationality isn't the issue
Sophos experts note all models can go wrong without governance — risk isn't unique to Chinese models.
In plain terms = open-source solves the 'who controls the off-switch' problem, but not the 'anyone can use it' problem — real security comes from monitoring mechanisms, not from a model's national label.
What does this mean for everyday users and businesses?
Le Chonk is free to use — enterprises and developers can deploy it without relying on Chinese models or paid closed-source alternatives.
But "free" does not mean "safe." Any organisation deploying an open-source model still needs its own usage monitoring and security governance processes.
This reflects a shift in the AI race: the central question is moving from "whose model is strongest" to "who can deploy models safely and under control."
市场有风险，内容仅供研究参考，不构成投资建议。