🔍 Read the full analysis: The AI Frontier Moves Ahead As Mistral Large 4 Lags on ThorstenMeyerAI.com

Prime Big Deal Days · Oct 6–7Offer from Amazon

Get office and shipping supplies delivered free — and shop member deals

• Fast, free delivery on millions of items
• Access to Prime Big Deal Days deals on October 6–7
• Prime Video, Amazon Music and more included

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Mistral AI introduced Mistral Large 4 in public API preview on October 6, but an October 7 comparison by ThorstenMeyerAI.com places its benchmark score below several leading U.S. and Chinese models. The model is Mistral’s largest to date, while its weights are not yet publicly downloadable; the company has scheduled their release for later in October.

Artificial Analysis assigns Mistral Large 4 Preview an Intelligence Index score of 38. In the comparison published by ThorstenMeyerAI.com, that matches OpenAI’s GPT-6 Luna at maximum reasoning effort, trails DeepSeek V4.1 Flash at maximum effort by one point, and sits below two other Chinese models and three leading U.S. models. The figures are a dated benchmark snapshot, not a direct prediction of success on any individual task.

The listed U.S. models are Anthropic Claude Opus 5.5 at 58, Google Gemini 4 Argon at 53, and OpenAI GPT-6.1 Sol at 52. China’s Z.ai GLM-5.3 scores 45 and Moonshot AI’s Kimi K3 scores 44. Cohere Command A+ of Canada scores 13, below Mistral. The comparison includes different reasoning settings and does not hold compute budgets constant; the source also cautions that developer locations do not indicate where API requests are processed.

Mistral describes Large 4 as a mixture-of-experts model with one trillion total parameters and 49 billion active parameters, accepting text and images. The company says it trained the model on its own European infrastructure and is continuing to improve it. Thorsten Meyer, the source article’s author, recommends against choosing the current preview for demanding agentic tasks, citing the benchmark results and his experience. His reported hallucinations are personal observations, not findings from a controlled comparison.

What the Score Means for Developers

The release gives developers a new Mistral model to evaluate, but the available evidence does not establish that it is a leading choice for complex, extended work. Agentic tasks involve multiple steps—planning, using tools, interpreting results and carrying decisions forward. Errors early in that process can shape later actions, so sustained accuracy and verification matter alongside a model’s headline capabilities.

The benchmark is relevant evidence, but it is not a test of every developer’s workflow. A score of 38 does not prove that Large 4 will fail a particular coding or research task. Likewise, a large context window—Artificial Analysis reports about 512,000 tokens—indicates how much input can fit, not whether the model will reason reliably over all of it. Developers considering the preview need task-specific testing before assigning it unsupervised work.

The comparison also has limits: reasoning settings differ, the scores are a snapshot, and the source says DeepSeek V4.1 Flash has approximately comparable benchmark intelligence at a much lower measured cost per task. The supplied material does not include the underlying cost figures, so the cost claim cannot be independently quantified here.

Preview Now, Weights Later

The distinction between an API preview and a public-weight release matters. As of October 7, Large 4 can be accessed through a preview API, but its weights are not yet available for public download. Mistral says those weights are due later in October, meaning the current decision for developers concerns the preview rather than a finished downloadable release.

Mistral’s launch is also a development in European AI capacity: the company says it trained the model on its own infrastructure in Europe. That fact describes where and how Mistral says it built the system; it does not establish that the model matches the strongest alternatives on performance. The cited source says Mistral is continuing to improve the preview, so subsequent versions and the eventual weight release may warrant separate evaluation.

“I would not choose it for demanding agentic work or long tasks when stronger models are available.”

— Thorsten Meyer, ThorstenMeyerAI.com

Performance Beyond the Benchmark

The available material does not provide controlled, workload-specific results showing how Large 4 performs on long agentic tasks, coding, or professional work. Mistral promotes strengths in agentic coding and specialized tasks, but those claims require testing against the workflows developers intend to use. The source author’s hallucination observations also cannot establish how often the model produces unsupported answers across users or tasks.

It is not yet clear whether later preview updates or the scheduled weight release will change the benchmark picture. The Artificial Analysis scores are a snapshot dated October 7, 2026, and the listed models were evaluated with different reasoning settings. The source provides no complete cost table, so its claim that DeepSeek is much cheaper cannot be assessed in detail from the supplied figures.

October Weights and Further Testing

Mistral has scheduled the model weights for release later in October, according to the source material. Developers and evaluators can then assess the released model and any preview updates against their own tasks, including whether it follows constraints, checks evidence, and maintains reliable performance across multiple steps.

Further benchmark results may also change the current comparison. Until those results and workload-specific evaluations are available, the evidence supports a limited conclusion: Large 4 is a new preview with substantial model scale, but the cited index places it behind several competing models, and the current material does not establish its reliability for demanding autonomous work.

Key Questions

Is Mistral Large 4 publicly available?

It is available as a public API preview, according to the source. Its model weights were not publicly downloadable as of October 7, 2026; Mistral has scheduled their release for later in October.

What score did Mistral Large 4 receive?

Artificial Analysis gives Mistral Large 4 Preview an Intelligence Index score of 38 in the comparison cited by ThorstenMeyerAI.com. The score is a benchmark result, not a percentage or a guarantee of performance on a specific task.

Which models scored higher in the cited comparison?

Anthropic Claude Opus 5.5 scored 58, Google Gemini 4 Argon 53, OpenAI GPT-6.1 Sol 52, Z.ai GLM-5.3 45, and Moonshot AI’s Kimi K3 44. The source notes that the comparison uses different reasoning settings.

Does the benchmark prove Large 4 will fail at agentic tasks?

No. The cited index does not directly measure reliability on every coding, research, or multi-step workflow. The source author advises against selecting the current preview for demanding agentic work, but frames that as a judgment informed by benchmarks and personal experience.

What remains to be evaluated?

Developers still need workload-specific tests of the preview and, when available, the public weights. The supplied material does not establish controlled hallucination rates, performance across long tasks, or detailed cost comparisons.

Source: ThorstenMeyerAI.com