🔍 Read the full analysis: Mistral Large 4 Compared: Still Behind 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 released Mistral Large 4 in public API preview on October 6, but its score of 38 on Artificial Analysis’s Intelligence Index leaves it behind leading US models and several Chinese competitors in the October 7 snapshot. The result matters to developers choosing models for complex work, though it does not by itself establish how the preview will perform on any particular workflow.

Mistral describes Large 4 as its largest model to date: a mixture-of-experts system with one trillion total parameters and 49 billion active parameters. It accepts text and images. The preview is currently available through an API; the model weights are not yet publicly downloadable. Mistral said it trained the model on its own infrastructure in Europe and plans to release the weights later in October.

Artificial Analysis’s October 7 comparison gives Large 4 a score of 38. In the same snapshot, Anthropic Claude Opus 5.5 scores 58, Google Gemini 4 Argon scores 53, and OpenAI GPT-6.1 Sol scores 52. Chinese models Z.ai GLM-5.3 and Moonshot AI’s Kimi K3 score 45 and 44, respectively; DeepSeek V4.1 Flash scores 39. OpenAI’s GPT-6 Luna also scores 38, while Canada’s Cohere Command A+ scores 13.

These are index points, not percentages or predicted task success rates. Artificial Analysis’s listed results use different named reasoning settings, so the models were not evaluated under identical compute budgets. The source author characterizes Mistral as behind leading US models and stronger Chinese alternatives, while noting that Large 4 matches GPT-6 Luna’s listed score.

Choosing Models for Long Tasks

The result is relevant to developers considering agentic workflows, where a model must plan, use tools, interpret results and maintain decisions over multiple steps. Errors early in a sequence can affect later work, and a fluent final response does not necessarily show that each step was sound. A benchmark score can inform model selection, but it cannot guarantee success or failure on a specific coding, research or business task.

The source article’s author says they would not select the current preview for demanding agentic work or long tasks when higher-scoring alternatives are available. That is an assessment by the author, not a finding established by the benchmark alone. They also report encountering hallucinations during personal use, but describe this as experience rather than a controlled comparison across models. Developers should test the preview against their own tasks and measure accuracy, supervision needs and cost before relying on it.

Mistral’s European infrastructure is relevant to its position as a European AI provider, but it does not establish that Large 4 matches the leading models on performance. The distinction for buyers is between the significance of the release for European capacity and the separate question of whether the preview is the right tool for a given workload.

Preview Now, Weights Later

The October 6 announcement opened API access to a preview rather than a finished public-weight release. Mistral says the weights are scheduled to arrive later in October and that it continues to improve the model. Results from the current version should be read as a dated snapshot, not a final assessment of later releases.

Artificial Analysis reports that Large 4 has a context capacity of roughly 512,000 tokens. That figure describes how much input the model can accept; it does not show whether the model will reason accurately across all of that material. Likewise, the index measures aggregate performance across a benchmark suite, not the reliability of every response or workflow.

The comparison’s developer locations refer to where the companies are based, not where an individual API request is processed. The source also cautions against claiming that every competitor is ahead: Cohere Command A+ scores below Mistral on this index. The narrower supported conclusion is that Large 4 trails the listed leading US models and several stronger-scoring Chinese alternatives.

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

— Thorsten Meyer, source article author

What the Preview Cannot Settle

The current benchmark snapshot does not show how Large 4 will perform after planned updates or after its weights are released. The weights are not yet available, and the source gives no results from a controlled test of the model against competitors on long agentic tasks.

The author’s hallucination observations are personal and do not establish comparative rates. It is also unclear from the provided material how the model performs across distinct professional workloads, or how users’ results may vary by prompting, tools and deployment. The available index score is not a direct measure of reliability, task completion or operating cost.

The source material’s cost discussion is incomplete: it ends before providing the figures. No cost-per-task comparison for Mistral Large 4 can be reported from the available information. DeepSeek V4.1 Flash is described by the source as having comparable measured benchmark intelligence at a much lower measured cost per task, but the underlying amounts and measurement details are not included.

Watch for Weights and Retests

Mistral has said it plans to release Large 4’s weights later in October 2026 and continues to improve the model. Those releases, and fresh independent evaluations, will provide a basis for checking whether its standing changes. Until then, the results describe an API preview rather than the planned public-weight version.

For developers, the practical next step is to evaluate the available preview on representative tasks and compare it with alternatives under clearly stated settings. Tests should track whether the model follows constraints, supports its conclusions, recovers from errors and completes work with an acceptable amount of human review. Those workload-specific results remain unavailable in the source material.

Key Questions

Is Mistral Large 4 publicly available?

It is available in public API preview, according to the source article. Its weights had not been made publicly downloadable as of October 7, 2026; Mistral said they were scheduled for release later in October.

How did Mistral Large 4 score?

Artificial Analysis gave the preview an Intelligence Index score of 38 in a snapshot dated October 7, 2026. The score is an index result, not a percentage or a prediction of success on a specific task.

Does the score prove the model will fail at agentic work?

No. The score is aggregate benchmark evidence, not a direct test of every long or tool-using workflow. The source author advises against choosing the preview for demanding agentic work, but that is their judgment, and they call for workload-specific testing.

What remains unconfirmed about its reliability?

The source provides no controlled comparative study of hallucination rates or long-task reliability. The author reports hallucinations in personal use and explicitly says that observation is not a controlled comparison.

Source: ThorstenMeyerAI.com