deepseek-v4-pro vs mistral-medium-2604
Pricing, Performance & Features Comparison
Context Length1M
Reasoning
Providers1
ReleasedApr 2026
Knowledge Cutoff–
LicenseMIT License
Flagship Mixture-of-Experts model with 1.6T total parameters and 49B activated per token. Trained on 32T+ tokens with hybrid attention for efficient 1M context processing.
Input$1.7
Output$3.5
Latency (p50)7s
Output Limit384K
Function Calling
JSON Mode
-
InputText
OutputText
in$1.7out$3.5cache$0.15write$1.7
Latency (24h)
Success Rate (24h)
Mistral’s current flagship (Medium 3.5): a frontier-class multimodal model optimized for agentic and coding work, succeeding Mistral Large 2.1, Devstral 2, and Magistral Medium.
Input$1.5
Output$7.5
Latency (p50)4s
Output Limit–
Function Calling
JSON Mode
-
InputText, Image
OutputText
in$1.5out$7.5––