🔍 Read the full analysis: Why Mistral Large 4 Still Trails The AI Frontier on ThorstenMeyerAI.com

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

Get tech for your team 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 Large 4 entered public API preview on October 6, but an Artificial Analysis Intelligence Index score of 38 puts it below several leading US and Chinese models in a snapshot published the next day. The result gives developers an early performance comparison, not a final verdict: Mistral says the model is still being improved, and its weights are not scheduled for release until later in October.

Artificial Analysis scores Large 4 Preview at 38 points, matching OpenAI’s GPT-6 Luna at maximum reasoning effort and sitting just below DeepSeek V4.1 Flash at 39. The same snapshot lists Z.ai’s GLM-5.3 at 45, Moonshot AI’s Kimi K3 at 44, OpenAI’s GPT-6.1 Sol at 52, Google’s Gemini 4 Argon at 53 and Anthropic’s Claude Opus 5.5 at 58. Cohere’s Command A+ scores 13. The figures are index points, not percentages or direct predictions of success on a particular task.

The comparison has limits: the models were assessed at different named reasoning settings, and the source says these are not evaluations under identical compute budgets. It is a dated snapshot that may change. Artificial Analysis reports that Large 4 has one trillion total parameters and 49 billion active parameters, accepts text and images, and has an approximately 512,000-token context window. Context capacity describes how much material can fit in a request; it does not demonstrate accurate reasoning across that material.

The source article’s author says they would not select the current preview for demanding agentic work or long tasks when stronger-scoring options are available. The author also reports encountering hallucinations in personal use, while explicitly describing that experience as not a controlled comparative study. Mistral, meanwhile, advertises strengths in agentic coding and specialized professional tasks. Those claims require testing on relevant workloads; the benchmark score alone does not establish how the model will perform in every developer’s workflow.

What the Score Means for Developers

For teams choosing a model to handle multi-step work, the comparison is an early signal about where Large 4 sits—not proof that it will fail a given project. Agentic systems plan, call tools, interpret results and carry decisions through successive steps. An unsupported assumption early in that chain can affect later actions, making reliability and verification as important as a model’s ability to accept a long prompt.

The article’s author argues that a lower aggregate score, alongside their own reported hallucinations, does not justify choosing this preview over substantially higher-scoring alternatives for complex autonomous work. That is a selection judgment, not a benchmark finding about all tasks. Developers should compare models on their own coding, research or professional workflows and account for the supervision each requires. The score also places a limit on what can be claimed for the launch: Mistral has made a large model available for testing, but the supplied evidence does not show parity with the highest-scoring competitors.

The market comparison is not uniformly unfavorable. Mistral scores above Canada-based Cohere’s Command A+ in this index, but that does not close the gap with the leading US models or the stronger Chinese alternatives in the table. Artificial Analysis’s location labels refer to developers, not where API requests are processed. The snapshot is useful as a comparison point, not as a complete assessment of deployment, privacy, cost or every model’s suitability.

Preview Status and Model Release

Mistral introduced Large 4 on October 6, 2026, as an API preview. The company says it trained the model on its own infrastructure in Europe and is continuing to improve it. The article describes the release as meaningful for European AI capacity, while distinguishing that development from evidence that the model is the strongest option for demanding work.

At the time of the October 7 analysis, the model’s weights were not publicly downloadable; their release was scheduled for later in the month. Developers assessing it then were evaluating an API preview, not a finished open-weight release. Future changes and the eventual availability of weights could affect how researchers and organizations test or deploy it, but those developments should not be treated as already delivered.

Artificial Analysis’s scores are a snapshot using the settings listed for each model. The source cautions that the comparisons do not use identical compute budgets. It also reports that DeepSeek V4.1 Flash has approximately comparable benchmark intelligence at a much lower measured cost per task, though the supplied material gives no price figures or detailed cost methodology. That cost claim should be considered separately from the index ranking and checked against a team’s own usage.

What the Preview Cannot Establish

The available comparison does not establish how Large 4 performs across specific coding, research or professional tasks, or how reliably it sustains long agentic workflows. The author’s hallucination observations are personal and were not collected through a controlled comparison; they do not provide a measured rate or show that other models do not hallucinate.

It is also unclear from the source material how the score may change as Mistral continues training and updating the preview, what results the model will achieve once its weights are released, or how it compares on the specific workloads developers care about. The listed reasoning settings differ, and Artificial Analysis says the models were not tested under identical compute budgets. The snapshot therefore supports a qualified ranking, not a definitive head-to-head conclusion.

Weights and Workload Tests Ahead

Mistral scheduled the Large 4 weights for release later in October 2026. Until then, developers can assess the announced preview through its API, while treating results as provisional. The next useful evidence will include updated benchmark results, independent workload-specific evaluations and direct tests of how often the model needs human checking during multi-step tasks.

Teams considering the model should compare it with alternatives on their own tasks and record accuracy, completion rates, tool-use errors, supervision needs and cost. Artificial Analysis’s snapshot can help identify candidates for those tests, but it cannot settle the choice by itself. Whether Large 4 closes the measured gap remains open pending further releases and evaluations.

Key Questions

What did Mistral announce?

Mistral introduced Large 4 as a public API preview on October 6, 2026. The model accepts text and images; its weights were scheduled for release later that month, not yet available at the time of the source analysis.

How does Mistral Large 4 score against other models?

Artificial Analysis gives Large 4 Preview an Intelligence Index score of 38 in its October 7 snapshot. That matches GPT-6 Luna at maximum reasoning effort, is just below DeepSeek V4.1 Flash at 39, and is below several listed US and Chinese models. The source cautions that settings and compute budgets differ.

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

No. An aggregate score is not a direct prediction for a particular workflow. The source author advises against choosing the current preview for demanding agentic tasks when stronger-scoring alternatives are available, but calls this a judgment about the preview; workload-specific testing is still needed.

Are claims about hallucinations based on a controlled test?

No. The source author reports encountering hallucinations in personal use and explicitly says this was not a controlled comparative study. The material provides no measured hallucination rate for Large 4 or its competitors.

When will Large 4’s weights be available?

Mistral scheduled the weights for release later in October 2026. The source material does not confirm that release has occurred, so their availability should be checked against a later announcement.

Source: ThorstenMeyerAI.com