🔍 Read the full analysis: Mistral Large 4’S Place In Today’s AI Frontier Race on ThorstenMeyerAI.com
Get smart everyday buys delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Mistral introduced Large 4 as an API preview on October 6, with a public release of its weights planned for later in October. Artificial Analysis rated the preview at 38 on its Intelligence Index, below several leading U.S. and Chinese models, while the source author reports concerns from personal use with hallucinations and long agentic tasks.
Mistral launched Large 4 in public API preview on October 6, introducing its largest model to date as the French company competes with leading AI developers. A benchmark snapshot from Artificial Analysis scores the preview at 38 on its Intelligence Index, below several U.S. and Chinese models; Mistral says the model’s weights are scheduled for release later in October.
Mistral describes Large 4 as a mixture-of-experts model with one trillion total parameters and 49 billion active parameters. It accepts text and images. The company said it trained the model on its own infrastructure in Europe and is continuing to improve it. At the time covered by the source material, the weights were not yet publicly downloadable, so the current release is an API preview rather than a completed open-weight release.
In Artificial Analysis’s October 7 snapshot, Large 4 Preview scored 38 on the Intelligence Index. The cited comparison lists Claude Opus 5.5 at 58, Gemini 4 Argon at 53 and GPT-6.1 Sol at 52. Chinese models GLM-5.3 and Kimi K3 scored 45 and 44, respectively; DeepSeek V4.1 Flash scored 39. GPT-6 Luna also scored 38, while Cohere Command A+ scored 13.
These are benchmark index points, not percentages or direct predictions of success on a particular task. The listed models were evaluated at different reasoning settings, and the comparison is a dated snapshot. The source author also reports encountering hallucinations while using the preview, but identifies that as personal experience, not a controlled comparison.
Frontier model watch · October 2026
Mistral Large 4’s Place in Today’s AI Frontier Race
Mistral’s new model arrives as a public API preview, with weights planned for later in October. Its benchmark score trails several listed frontier models, while key questions about real-world reliability and the final release remain open.
01 / Model profile
What Mistral has introduced
Large 4 is Mistral’s largest model to date, designed as a mixture of experts. The current offering is an API preview; the planned weights were not publicly downloadable in the source snapshot.
Architecture
Mixture of experts
One trillion total parameters, with 49 billion active parameters. Mistral describes it as its largest model so far.
Inputs
Text + images
The model accepts text and image inputs. Artificial Analysis reports roughly 512,000 tokens of context capacity.
Infrastructure
Trained in Europe
Mistral says it trained Large 4 on its own European infrastructure and continues to improve the model.
02 / Benchmark snapshot
A visible gap at the top
Artificial Analysis’s October 7 Intelligence Index comparison places Large 4 Preview below several listed U.S. and Chinese models.
03 / Reading the evidence
What the score can—and cannot—tell you
A broad benchmark can frame a comparison. It cannot establish reliability across your specific coding, research, or professional workflow.
An early mistaken assumption can affect later tool calls and decisions. Sustained accuracy and verification matter.
The author’s reported hallucinations are anecdotal, not a measured rate or controlled comparison.
The source material cuts off during its cost discussion, so it does not establish a full price comparison.
04 / Practical path
From preview to informed choice
Test the model against the work you need it to do, then revisit the evidence as the release and independent evaluations develop.
Try your tasks
Use representative coding, research, or professional examples.
Check support
Verify claims, sources, and whether constraints are followed.
Stress multiple steps
Look for errors that compound across plans, tools, and handoffs.
Reassess later
Compare updated benchmarks and the eventual weights when available.
05 / Attributed perspectives
Launch ambition, evaluation still ahead
Mistral’s infrastructure milestone and the model’s suitability for demanding work are separate questions.
“Mistral said it trained the model on its own infrastructure in Europe and continues to improve it.”Mistral · Company statement
“I would not choose it for demanding agentic work or long tasks when stronger models are available.”Thorsten Meyer · ThorstenMeyerAI.com
The recommendation is the source author’s assessment based on the preview, benchmark snapshot, and personal use—not a universal finding established by the index.
06 / Key questions
What to know now
Release status and performance claims refer to the source material’s October 2026 snapshot.
Has Mistral Large 4 been released?
It was announced as a public API preview on October 6, 2026. The weights were planned for later in October but were not yet publicly downloadable in the source material.
How did it score?
Artificial Analysis gave Large 4 Preview 38 points on October 7. The comparison uses different reasoning settings and is not a percentage or task-specific prediction.
Does 38 mean it will fail at agentic work?
No. The index does not test every workflow. The author recommends starting with stronger-scoring alternatives for complex autonomous work, an assessment rather than a universal finding.
What remains unknown?
The final performance and timing of the planned weights, controlled comparisons of hallucination rates, and a complete price comparison were not established.
Benchmark Gaps Shape Model Choices
The scores matter to developers weighing Large 4 for demanding work, but they do not settle every use case. A model’s aggregate benchmark result can help frame comparisons; it cannot establish how reliably that model will perform on a specific company’s coding, research or professional workflow. In particular, the source author argues that the available evidence does not support choosing this preview over higher-scoring alternatives for long agentic tasks.
Agentic systems carry plans through multiple steps, often using tools and acting on earlier results. A mistaken assumption early in a workflow can affect later decisions, so sustained accuracy and verification matter alongside the ability to accept a large amount of input. Artificial Analysis reports roughly 512,000 tokens of context capacity for Large 4, but context capacity describes how much material can fit in a request; it does not establish accurate reasoning across that material.
The launch also has a broader European dimension. Mistral says it trained the model on its own European infrastructure, a development relevant to the region’s AI capacity. That achievement and the model’s commercial suitability are separate questions: the benchmark snapshot places it below several leading competitors, while the preview and planned weight release remain subject to change.
Products Worth Considering
FLEXISPOT 36-Inch Height-Adjustable Standing Desk Converter
Dell XPS 13 9350 Copilot+ Laptop, 13.4-inch 120Hz Display, Intel Core Ultra 7 256V, 16GB RAM, 512GB SSD, Ubuntu Linux 24.04 LTS
FLEXISPOT Standing Desk with Monitor Riser & Cable Management
As an affiliate, we earn on qualifying purchases.
A Preview, Not a Finished Release
Mistral announced the preview API on October 6. Its planned later-October weight release had not happened at the time described in the source material. Readers should distinguish that schedule from a release already available for download, and treat performance claims about the preview separately from future versions that Mistral may improve.
The benchmark table compares developers’ locations, not where a particular API request is processed. It also combines different named reasoning settings rather than evaluations under identical compute budgets. Artificial Analysis’s Intelligence Index offers one aggregate comparison, but the source author notes that a score of 38 is not a direct measure of reliability on an individual coding or research task.
The source author’s recommendation is an assessment, not a finding established by the benchmark: based on the current preview, benchmark results and personal use, the author would begin with stronger-scoring alternatives for complex autonomous work. Mistral’s advertised strengths in agentic coding and specialized professional tasks may still matter, but those claims call for testing on the workloads in question.
“Mistral said it trained the model on its own infrastructure in Europe and continues to improve it.”
— Mistral
Preview Results Leave Questions Open
Performance may change as Mistral continues development and releases the weights. The source material does not establish the final performance of that release or whether it will match the preview API. It also does not give a controlled, task-by-task comparison showing how Large 4 performs against the listed alternatives in real agentic workflows.
The author’s reported hallucinations reflect personal experience, not a measured rate or a systematic comparison across models. The available material also cuts off during its discussion of costs, so it does not provide enough information to assess a full price comparison. The benchmark scores are dated, use different reasoning settings and may change as models and evaluations are updated.
Watch for Weights and Testing
The next stated milestone is Mistral’s planned release of Large 4’s weights later in October. Availability and final timing were not confirmed in the source material. Once released, developers will be able to assess the weights as well as the API preview, subject to Mistral’s release terms.
For now, the practical next step for prospective users is to test the preview on their own tasks, checking whether outputs are supported, constraints are followed and work remains reliable across multiple steps. Updated benchmarks and broader independent evaluations may clarify how the model compares after further development; the October 7 snapshot should not be treated as a permanent ranking.
Key Questions
Has Mistral Large 4 been released?
It is available as a public API preview, announced October 6, 2026. The source material says its weights were scheduled for release later in October, but were not yet publicly downloadable at the time.
How did Large 4 score against other models?
Artificial Analysis gave Large 4 Preview an Intelligence Index score of 38 in a snapshot dated October 7. That was below several listed U.S. and Chinese models. The scores use different reasoning settings and are not percentages or predictions of success on a specific task.
Does a score of 38 mean Large 4 will fail at agentic work?
No. The index is an aggregate benchmark, not a direct test of every coding, research or agentic workflow. The source author considers the score insufficient reason to choose the preview over higher-scoring alternatives for complex autonomous work, but that is an assessment rather than a universal finding.
What remains unknown about the release?
The final performance and availability of the planned weights were not confirmed in the source material. It also does not provide a controlled comparison of hallucination rates or enough cost data to assess the model’s price against alternatives.
Source: ThorstenMeyerAI.com
