Standard Compute
Model comparison 2026

Devstral 2 vs Step 5 Preview

Devstral 2
Mistral · Balanced · Open weights
$0.88/1M · 262K ctx
Step 5 Preview
StepFun · Balanced · Proprietary
$1.51/1M · 1M ctx

Community vote

Pick a winner in each dimension. You can change your vote any time.

Output Quality
Correctness and depth of what it produces
Agentic Ability
Tool calls, instruction-following, and multi-step tasks
Speed
Tokens per second and time-to-first-token
Value for money
How much capability you get per dollar
Reliability
Consistent results: fewer refusals, loops and format breaks

Specs and pricing side by side

Spec
Devstral 2
Step 5
Maker
Mistral
StepFun
Blended price / 1M
$0.88
$1.51
Input / output
$0.40 in · $2.00 out / 1M tokens
$1.00 in · $2.70 out / 1M tokens
Context window
262K
1M
Open weights
Yes
No
Tool use
Yes
Yes
Reasoning
No
Yes
Output Quality
3.3
8.2
Agentic Ability
4.4
7.6
Speed
8.2
7.5
Value for money
5.7
7.1
Reliability
8.4
8.6

Pricing and capabilities synced from the OpenRouter catalogue. Scores are editorial (0–10).

Verdict: Devstral 2 or Step 5?

Updated 2026-10-13

Choose Devstral 2 if you want coding agents that need a Mistral open-weight model. Choose Step 5 Preview if you want mid-priced agent loops that want more capability than a Flash-class model.

In our editorial scoring, Step 5 Preview leads in 4 of five dimensions (output quality, agentic ability, value for money and reliability), while Devstral 2 leads in 1. On price, Devstral 2 runs about $0.88 per 1M tokens (blended) and is open-weight; Step 5 Preview is about $1.51 and proprietary.

Where Devstral 2 falls short
  • 0% in Artificial Analysis's Terminal-Bench 4.0 run
  • 262K context, smaller than 11.7% of the coding-agent requests we measured
Full Devstral 2 breakdown →
Where Step 5 falls short
  • Still a preview, so behavior and price can change
  • Only StepFun serves it on OpenRouter, so there is no provider fallback
Full Step 5 breakdown →

The agent matters as much as the model. Compare AI agents →

Current Claude, GPT, Gemini and open models are included with Standard Compute

Standard Compute

Current frontier models Claude Opus 5.5 · Claude Sonnet 5.5

Your agents never stop.
Your bill never grows.

Frontier models when it counts. Efficient models when it doesn’t. The most intelligence per dollar. One flat bill.

Plans from $19/mo · cancel anytime · 7-day fair refund

7-day fair refund No credit card needed Billing by Stripe1,800+ agent users

Related comparisons