HomeAI models
Mistral: Mistral Large
Mistral AI
•
Mistral Large (Feb '24)
•
Released 
November 2024

Mistral AI
Mistral Large (Feb '24)

Mistral's 123B dense flagship for function-calling agents, multilingual generation and code across 80+ programming languages.

Modality:
Text
PDF
model ID
mistralai/mistral-large
Start for free
arrow black icon
View API Docs
arrow white icon

Output Speed *

0.00
tok/s

Intelligence Index *

5.8
/ 100

Context Window *

128000
tokens

Input price

12
Anytoken

Output price

36
Anytoken

Mistral Large: A 123B Dense Model Built for Function Calling and Multilingual Code
‍

Mistral Large is Mistral AI's 123-billion-parameter dense flagship, positioned at the top of the pre-Mistral-Large-3 lineup. The November 2024 (24.11) update improved function-calling accuracy, system-prompt handling and long-context behaviour over the July 24.07 release. It is a text-only model trained on 80+ programming languages and dozens of natural languages. Its strongest fit is agentic workloads that need native tool calling, reliable JSON output and consistent multilingual generation, rather than the highest possible reasoning scores now offered by newer reasoning models.

Integrate Mistral Large via the AnyAPI.ai API

Performance

Where Mistral Large Earns Its Place: Code, Tools and Languages

Mistral Large is built around agentic and coding work rather than frontier reasoning. Mistral trained it on a large proportion of code across 80+ programming languages, and independent aggregations report roughly 92% on HumanEval and 93% on GSM8K, with an 84% MMLU score on the pretrained base. Practically, this makes it dependable for generating and refactoring code and for driving tool-using workflows through native function calling and JSON output. The consequence for production teams: it slots in as a capable general-purpose backend for structured, multilingual tasks—though newer reasoning models now lead on hard math and analytical benchmarks.

Benchmarks

Mistral Large on Independent Benchmarks and Long Context

Independent aggregations of Mistral Large 2 report about 92% on HumanEval, 93% on GSM8K (8-shot) and 84% on MMLU, placing it near frontier open models of its 2024 generation. On Artificial Analysis' composite index, however, Mistral Large 2411 sits well below current reasoning models, reflecting its non-reasoning, 2024-era design. Long-context evaluations such as LongBench Pro show it holding coherence across 128K-token inputs but trailing larger dense models like Llama 3.1 405B. Interpretation: strong for its era on code and instruction following, no longer competitive on the hardest reasoning tasks.

Output Speed

*
0.00
tok/s

Intelligence Index

*
5.8
/ 100

MMLU *

Broad world knowledge and problem-solving
52
%

GPQA *

PhD-level scientific reasoning across physics, biology, chemistry.
35
%

HLE *

Adherence to multi-step structured instructions.
4
%

LiveCodeBench *

Tool-calling reliability in long agentic loops.
18
%

Technical Specifications

What the model supports

Mistral Large is a text-in, text-out dense model with a 128,000-token context window and 123B parameters. It supports native function calling, JSON structured outputs and streaming, and the 24.11 update added improved system-prompt handling. It is not a reasoning model—there is no extended-thinking mode or reasoning-effort control. The most production-relevant point is modality: unlike sibling Pixtral Large, Mistral Large does not accept image input. Teams needing vision or document-image understanding must pair it with a multimodal model rather than expecting Mistral Large to handle those inputs directly.
Verified Specifications — 
Mistral Large (Feb '24)
*
Input modalities
Text
PDF
output modalities
Text
Context window
128000
 tokens
Maximum output tokens
102400
Reasoning
No
Knowledge cutoff
November 2024
Pricing (standard)
12
 AnyTokens in
 / 
36
 AnyTokens out

Quickstart

Sample code for Mistral Large (Feb '24)

Python
JavaScript
Curl
import requests‍

url = "https://api.anyapi.ai/v1/chat/completions"
payload = {    
  "messages": [        
    {            
      "content": "test prompt",           
      "role": "user"        
    }    
  ],    
  "model": "mistral-large"
  }
headers = {    
  "Authorization": "Bearer  AnyAPI_API_KEY",    
  "Content-Type": "application/json"
}

response = requests.post(url, json=payload, headers=headers)
print(response.json())‍
import requests url = "https://api.anyapi.ai/v1/chat/completions" payload = { "messages": [ { "content": "\"test prompt\"", "role": "user" } ], "model": "mistral-large" } headers = { "Authorization": "Bearer AnyAPI_API_KEY", "Content-Type": "application/json" } response = requests.post(url, json=payload, headers=headers) print(response.json())
View docs
Copy
Code is copied
const url = 'https://api.anyapi.ai/v1/chat/completions';
const options = 
{  
method: 'POST',  
headers: {
  Authorization: 'Bearer  AnyAPI_API_KEY', 
  'Content-Type': 'application/json'
},  
body: '{
  "messages":[
    {
      "content":"test prompt",
      "role":"user"
    }
  ],
  "model":"mistral-large"
  }
'};
try {  
  const response = await fetch(url, options);  
  const data = await response.json();  
  console.log(data);
} 
catch (error) {  
  console.error(error);
}
const url = 'https://api.anyapi.ai/v1/chat/completions'; const options = { method: 'POST', headers: {Authorization: 'Bearer AnyAPI_API_KEY', 'Content-Type': 'application/json'}, body: '{"messages":[{"content":"\"test prompt\"","role":"user"}],"model":"mistral-large"}' }; try { const response = await fetch(url, options); const data = await response.json(); console.log(data); } catch (error) { console.error(error); }
View docs
Copy
Code is copied
curl --request POST \  
  --url https://api.anyapi.ai/v1/chat/completions \  
  --header 'Authorization: Bearer  AnyAPI_API_KEY' \  
  --header 'Content-Type: application/json' \  
  --data '{  
  "messages": [    
    {      
      "content": "test prompt",      
      "role": "user"    
    }  
  ],  
  "model": "mistral-large"
  }'
curl --request POST \ --url https://api.anyapi.ai/v1/chat/completions \ --header 'Authorization: Bearer AnyAPI_API_KEY' \ --header 'Content-Type: application/json' \ --data '{ "messages": [ { "content": "\"test prompt\"", "role": "user" } ], "model": "mistral-large" }'
View docs
Copy
Code is copied
View docs
Code examples coming soon...

Limitations & Trade-offs

Where Mistral Large (Feb '24) falls short

1
No vision or multimodal input. Mistral Large accepts text only; its sibling Pixtral Large is the multimodal option. Any workload involving images, charts, screenshots or scanned documents cannot be served by Mistral Large directly, forcing you to add a separate multimodal model. For document-heavy pipelines this is the most common reason to look elsewhere within Mistral's lineup.
2
Not a reasoning model. Mistral Large has no extended-thinking mode or reasoning-effort control, and independent composite indexes place it well below current reasoning models on hard math and analytical benchmarks. For step-heavy problem solving, competitive programming or research-grade reasoning, a dedicated reasoning model will outperform it despite Mistral Large's solid instruction following.
3
2024-era generation, superseded internally. The model belongs to the pre-Mistral-Large-3 lineup, and the earlier 24.07 release is already marked retired in Mistral documentation. Teams building for long-term production should confirm availability and weigh whether a newer Mistral flagship offers better intelligence-per-cost before committing.
4
Self-hosting is heavy and non-commercial. The open weights are released under the Mistral Research License, which does not permit commercial use, and running the 123B dense model requires over 300 GB of GPU memory. Commercial self-hosting is therefore impractical for most teams—API access is the realistic path to production.

Best-Fit Workloads

Where this model earns its place

01

Function-calling agents

‍
The 24.11 update specifically improved function-calling accuracy and system-prompt handling, and the model outputs valid JSON natively. This makes it well suited to multi-step agents that decide which external tools or APIs to invoke and return structured parameters without hallucinating function signatures. It fits orchestration layers where reliable tool selection matters more than deep reasoning.

02

Multilingual code generation
‍

Trained on 80+ programming languages and dozens of natural languages, Mistral Large reports strong code-generation results (around 92% HumanEval in independent aggregations). It suits assistants that generate, explain or refactor code and must also communicate with users in French, German, Spanish, Chinese, Japanese and other languages within one model.

03

Structured multilingual content
‍

With native JSON output and broad language coverage, the model is a fit for pipelines that extract or generate structured data from multilingual text—product catalogs, localized content, or form-filling. Its instruction-following was a deliberate focus area, helping keep outputs concise and schema-conformant across languages.

04

Long-document processing
‍

The 128K-token context window lets Mistral Large ingest large documents or codebases in a single request. Long-context evaluations show it maintaining coherence across extended inputs, though larger dense models score higher. It suits summarization and analysis of long text where consistent instruction following outweighs the need for frontier reasoning.

Pricing in anytokens via AnyAPI
Input
12
₳
Output
36
₳
Cache write
—
₳
Cache read
—
₳
Get API access
arrow black icon
Read Docs

on this page

PerformanceBenchmarksTechnical SpecificationsQuickstartLimitations & Trade-offsBest-Fit WorkloadsIntegrationFAQ

Integration

Access Mistral Large (Feb '24) via AnyAPI.ai

Access Mistral Large (Feb '24) through AnyAPI.ai using a unified API built for multi-model AI applications. Integrate Mistral Large (Feb '24) without maintaining a separate provider-specific connection, and keep the flexibility to test, switch, or combine models as your application requirements evolve.

01

One API integration

Access Mistral Large (Feb '24) and other AI models through the same API workflow instead of maintaining separate integrations for every provider.

02

Easy model switching

Test Mistral Large (Feb '24) against alternative models or switch models as your performance, capability, or cost requirements change without rebuilding your application around another provider API.

03

Flexible for production

Use Mistral Large (Feb '24) from experimentation through production while keeping your AI stack flexible as workloads, traffic, and model requirements evolve.

04

Multi-model applications

Use Mistral Large (Feb '24) for the workloads where it performs best and combine it with other models for tasks that require different capabilities, performance, or efficiency.

Start free
arrow black icon
Read Docs

Frequently Asked Questions

Answers to common questions about integrating and using this AI model via AnyAPI.ai

Mistral Large supports a 128,000-token context window, shared across both the July 2024 (24.07) and November 2024 (24.11) releases. This is a substantial increase over the original 24.02 Mistral Large, which had a 32,000-token window, and it lets the model process long documents or large code files in a single request.

No. Mistral Large is a text-only model for both input and output. Mistral's multimodal sibling, Pixtral Large, released alongside the November 2024 update, is the model that accepts image input. If your workload requires interpreting images, charts or scanned documents, use Pixtral Large or another multimodal model rather than Mistral Large.

No. Mistral Large is a standard instruction-tuned model with no extended-thinking mode or reasoning-effort control. It performs well on code generation and instruction following, but on hard math and multi-step analytical benchmarks it trails dedicated reasoning models. Choose a reasoning model for research-grade or step-heavy problem solving.

Mistral Large was trained on a large proportion of code across 80+ programming languages. Independent aggregations report around 92% on HumanEval and strong results on related coding benchmarks, making it a capable choice for generating, explaining and refactoring code, especially in multilingual assistants.

The open weights are released under the Mistral Research License, which permits research and non-commercial use only. Running the 123B dense model also requires over 300 GB of GPU memory. For commercial production, API access is the practical path rather than self-hosting.

* Benchmark data source: Artificial Analysis artificialanalysis.ai