mistral-small-2603 vs grok-4.20-0309-non-reasoning

Pricing, Performance & Features Comparison

mistral-small-2603

Authormistral

Context Length256K

Reasoning

Providers1

ReleasedMar 2026

Knowledge Cutoff–

License–

Mistral Small 4: a hybrid Apache 2.0 model consolidating Magistral reasoning, Devstral coding agents, and Mistral Small into one model with toggleable reasoning effort.

Input$0.15

Output$0.6

Latency (p50)3.8s

Output Limit–

Function Calling

JSON Mode

InputText, Image

OutputText

mistral

in$0.15out$0.6––

Latency (24h)

Success Rate (24h)

grok-4.20-0309-non-reasoning

Authorxai

Context Length1M

Reasoning

Providers1

ReleasedMar 2026

Knowledge Cutoff–

License–

High-throughput non-reasoning variant of grok-4.20.

Input$1.3

Output$2.5

Latency (p50)4s

Output Limit–

Function Calling

JSON Mode

InputText, Image

OutputText

xai

in$1.3out$2.5cache$0.2–