Magica logo

Codestral 2508 vs DeepSeek V4 Pro (Comparative Analysis)

Loading comparison form...
Comparative Analysis: Codestral 2508 vs. DeepSeek V4 Pro
Want to try out these models side by side?Try Magica for free

Overview

Codestral 2508 was released 8 months before DeepSeek V4 Pro.
Codestral 2508Codestral 2508
DeepSeek V4 ProDeepSeek V4 Pro
Model Provider
The organization behind this AI's development
Mistral logoMistral
DeepSeek logoDeepSeek
Input Context Window
Maximum input tokens this model can process at once
256K
tokens
1M
tokens
Output Token Limit
Maximum output tokens this model can generate at once
Not specified
tokens
384K
tokens
Release Date
When this model first became publicly available
August 1st, 2025
April 24th, 2026
Knowledge Cutoff
Latest training-data date reported by the provider
March 31, 2025
Not reported

Capabilities & Features

Compare supported features, modalities, and advanced capabilities
Codestral 2508
DeepSeek V4 Pro
Input Types
Supported input formats
📝Text📁File
📝Text
Reasoning Controls
Defaults and configurable reasoning effort
Not reported
OptionalEffort: Extra High, HighDefault: High
Output Types
Supported output formats
📝Text
📝Text
Tokenizer
Text encoding system
MistralDeepSeek
Key Features
Advanced capabilities
Function CallingStructured OutputReasoning ModeContent Moderation
Function CallingStructured OutputReasoning ModeContent Moderation
Open Source
Model availability
ProprietaryAvailable on HuggingFace →

Pricing

Codestral 2508 is roughly 0.7x less expensive compared to DeepSeek V4 Pro for input tokens and roughly 1.0x more expensive for output tokens.
Codestral 2508Codestral 2508
DeepSeek V4 ProDeepSeek V4 Pro
Input Token Cost
Cost per million input tokens
$0.30
per million tokens
$0.43
per million tokens
Output Token Cost
Cost per million output tokens
$0.90
per million tokens
$0.87
per million tokens
Cache Read Cost
Cost to reuse cached input tokens
$0.03
per million tokens
$0.0036
per million tokens

Benchmarks

Compare relevant benchmarks between Codestral 2508 and DeepSeek V4 Pro.
Codestral 2508Codestral 2508
DeepSeek V4 ProDeepSeek V4 Pro
Intelligence Index
Overall model quality across independent evaluations
Benchmark not available.
44.3
(Artificial Analysis index; higher is better)
Coding Index
Programming performance across independent evaluations
Benchmark not available.
59.4
(Artificial Analysis index; higher is better)
Agentic Index
Ability to complete multi-step agentic tasks
Benchmark not available.
36.4
(Artificial Analysis index; higher is better)
Best Design Arena Score
Highest human-preference Elo score across design arenas
1,078 Elo
(#84 in 3D (45.5% win rate))
1,320 Elo
(#9 in 3D (60.1% win rate))

At a Glance

Quick overview of what makes Codestral 2508 and DeepSeek V4 Pro unique.
Mistral logoCodestral 2508 by Mistral can use external tools and APIs, generates structured data. It can handle standard conversations with its 256K token context window. Very affordable at $0.30/M input and $0.90/M output tokens. Released August 1st, 2025.
DeepSeek logoDeepSeek V4 Pro by DeepSeek can use external tools and APIs, offers advanced reasoning, generates structured data. It can handle standard conversations with its 1M token context window. Very affordable at $0.43/M input and $0.87/M output tokens. Released April 24th, 2026.

Explore More Comparisons

Compare your models with top performers across different categories