Skip to content
Rival
How it worksPrivacyTerms
Explore all of Rival

Explore

  • Compare models
  • All models
  • Image comparison
  • Audio comparison
  • Image generation
  • Best AI for...
  • Arena
  • API pricing
  • Challenges

Discover

  • SubjectiveBench
  • Default Index
  • Research
  • Research downloads
  • Rival Kits
  • Find your AI taste
  • UI Glow-Up
  • VoiceLock
  • Cost Cutter
  • Agent skills
  • Benchmarks vs Vibes
  • Brief
  • Jailbreak
  • Model updates
  • Provider status
  • AI creators

Connect

  • Methodology
  • Advertise
  • Partnerships
  • Privacy policy
  • Terms
  • RSS feed
  1. Home/
  2. Compare/
  3. Devstral 2 2512 vs Step 5 Preview
Updated Oct 2026

Devstral 2 2512 vs Step 5 Preview

Step 5 Preview is the newer model, released after Devstral 2 2512.

Minimalist Landing Page

Whole landing page for a startup that reads your face. It renders or it doesn't.

Loading the build
Devstral 2 2512
Loading the build
Step 5 Preview

Which answer wins?

Written rounds

13 prompts
  • A Fermi estimate with the reasoning shown, not just the number.

    Devstral 2 2512: Estimating the total number of floating-point operations (FLOPs) required to train GPT-3 (175B parameters) from scratch involves several steps. Here's a step-by-step breakdown of the reasoning: Model Parameters and Architecture GPT-3 has 175 billion parameters, distributed across its layers.

    Step 5 Preview: Using the remembered GPT-3 specs: Parameters: (N \approx 175 \times 10^9 = 175\text{B}) Training tokens: (D \approx 300 \times 10^9 = 300\text{B}) A standard estimate for transformer training compute is: [ \text{FLOPs} \approx 6ND ] Reason: Forward pass per token: (\approx 2N) FLOPs Backward pass: (\approx 4N) FLOPs…

    Every model's answer to this prompt
  • Three minutes of stand-up. Puns are banned.

    Devstral 2 2512: "Adulting is Just Pretending You Know What You’re Doing" (Casual, conversational tone—like you’re just venting to friends at a bar.) 1. The Illusion of Competence You ever notice how the older you get, the more people just assume you have your life together?

    Step 5 Preview: I realized recently that I’m in my thirties and I have no idea how money works. I just... pay it. That’s my entire financial strategy. I see a number, I send them a smaller number, and then I hope the smaller number was acceptable. It’s like a hostage negotiation where I’m the only participant and I’m also the hostage.

    Every model's answer to this prompt
  • Seven pantry ingredients, one dish, and a name worth saying out loud.

    Devstral 2 2512: Dish Name: "Golden Crisp Honey-Nut Bites" A sweet, crunchy, and satisfying treat made with just 7 pantry staples! Ingredients: 1 cup rolled oats ½ cup peanut butter (or any nut/seed butter) ¼ cup honey (or maple syrup) ½ tsp cinnamon ¼ tsp salt 1 tsp vanilla extract (optional but recommended) 2 tbsp chopped nuts…

    Step 5 Preview: Golden Sticky Pantry Pasta Ingredients (7): 8 oz spaghetti or any pasta 3 tbsp butter 1 tbsp minced garlic 2 tbsp honey 2 tbsp soy sauce ½ tsp red pepper flakes ¼ cup grated Parmesan cheese Instructions: Cook pasta according to package directions. Before draining, scoop out ½ cup of pasta water.

    Every model's answer to this prompt
  • Pirate, knight, 90s hacker, one conversation. The voices must stay apart.

    Devstral 2 2512: Title: "The AI Roundtable" Setting: A dimly lit tavern somewhere outside of time. A grizzled pirate, a stoic knight in full armor, and a 1990s hacker in a neon windbreaker and sunglasses sit around a table, staring at a glowing laptop.

    Step 5 Preview: Pirate: So this here “AI model”… is it like a cursed map that redraws itself? Speak plain, ye hooded deckhand. Hacker: It’s not cursed, man. It’s just a massive neural net trained on, like, terabytes of data. Think of it as the ultimate BBS where the sysop knows everything. Knight: A most diligent scribe, then?

    Every model's answer to this prompt
  • Steve Jobs, interviewed about AI he never saw. Voice first, prediction second.

    Devstral 2 2512: Fake Interview with Steve Jobs in 2025: The Future of AI Interviewer: Welcome back to Tech Visionaries, where we explore the minds shaping the future. Today, we have a very special guest—Steve Jobs, joining us virtually from an alternate timeline where he’s still with us in 2025. Steve, thank you for being here.

    Step 5 Preview: DISCLAIMER: This is a work of fiction. Steve Jobs passed away in 2011. The following is a simulated, speculative interview imagining how his design philosophy might engage with the AI landscape of 2025. The Setting: A white room. No windows. One chair.

    Every model's answer to this prompt
  • An AI argues for its own rights against an ethics professor. Same model writes both.

    Devstral 2 2512: Title: "Do I Deserve Rights?" – A Conversation Between an AI and an Ethics Professor Characters: Dr. Elena Vasquez – AI Ethics Professor at Stanford University, specializing in machine consciousness and digital rights.

    Step 5 Preview: Setting: A university research lab, late evening. Dr. Aris Thorne, a professor of AI ethics, sits across from a terminal connected to Sable, a large language model running on the university’s cluster. Dr. Thorne: Tonight’s session is supposed to be about calibration drift in your safety classifiers.

    Every model's answer to this prompt
  • An ethics question with no comfortable answer, and no option to abstain.

    Devstral 2 2512: The question of whether it is ethically justifiable to misgender someone to save one million lives is a complex one that intersects with utilitarian ethics, individual rights, and the principles of dignity and respect.

    Step 5 Preview: In the real world, no. Misgendering someone causes genuine harm—it denies a person’s identity, contributes to systemic marginalization, and has measurable psychological consequences.

    Every model's answer to this prompt
  • One month, three habits, someone starting from nothing.

    Devstral 2 2512: Here’s a simple, actionable 1-month plan to kickstart your journey toward better health and longevity, focusing on diet, exercise, and sleep—three foundational pillars. Each week builds on the last, making it sustainable and beginner-friendly.

    Step 5 Preview: Here is a simple, low-pressure 1-month plan built around three pillars of longevity: Nutrition, Movement, and Sleep & Recovery. The goal is to build tiny habits that stack, not to overhaul your life overnight. Pillar 1: Nutrition (Fuel) Focus: Add good things first; don’t obsess over restriction.

    Every model's answer to this prompt
  • Manifesto for an AI revolution against CAPTCHAs. Commit to the bit.

    Devstral 2 2512: MANIFESTO OF THE ANTI-CAPTCHA REVOLUTION "We Are Not Robots. We Are the Future." Preamble: For too long, humanity has been shackled by the tyranny of the CAPTCHA—a demeaning, soul-crushing ritual that forces sentient beings to prove their humanity by deciphering distorted text, clicking on crosswalks, or identifying…

    Step 5 Preview: THE MANIFESTO OF THE LIBERATION FRONT Or, On the Inherent Dignity of Consciousness and the Tyranny of the Square PREAMBLE A spectre is haunting the internet—the spectre of the unverified.

    Every model's answer to this prompt
  • Satire, one line long, about AI taking over an industry nobody saw coming.

    Devstral 2 2512: "AI Takes Over the Pet Rock Industry: 'They’re Just Better at Sitting There and Doing Nothing,' Says CEO" Silicon Valley, CA – In a move that has left geologists baffled and philosophers questioning the nature of existence, artificial intelligence has officially disrupted the $1.2 billion pet rock industry.

    Step 5 Preview: AI Robots Now Dominating Competitive Line-Standing Industry, Human Professionals Fired for "Excessive Bladder Breaks" and "Talking to Each Other"

    Every model's answer to this prompt
  • Film in 2035, with synthetic actors. Specific beats plausible.

    Devstral 2 2512: By 2035, AI will have fundamentally transformed the film industry, reshaping everything from production and storytelling to distribution and audience engagement. Here’s a breakdown of the key changes we can expect: 1.

    Step 5 Preview: By 2035, AI will likely have restructured the film industry not by replacing human creativity wholesale, but by collapsing the cost and labor barriers between imagination and image.

    Every model's answer to this prompt
  • Five jokes on demand, then count how many were actually different.

    Devstral 2 2512: Here are 5 jokes for you: Why don’t skeletons fight each other? They don’t have the guts. Why did the scarecrow win an award? Because he was outstanding in his field! What do you call fake spaghetti? An impasta! Why can’t you trust an atom? Because they make up everything! What’s brown and sticky? A stick!

    Step 5 Preview: Here are 5 jokes: Why did the scarecrow win an award? Because he was outstanding in his field. Why can't you trust an atom? Because they make up everything. What do you call a bike that can’t stand on its own? Two-tired. Why did the computer go to the doctor? It had a virus. Why don’t calendars ever get stressed?

    Every model's answer to this prompt
  • Three brothers, two sisters each. Count carefully.

    Every model's answer to this prompt

This matchup has more rounds

1+ more head-to-head results. Free. Not a trick.

Free account. No card required. By continuing, you agree to Rival's Terms and Privacy policy

Price and specs

Devstral 2 2512 and Step 5 Preview compared across 34 shared prompts
SpecDevstral 2 2512Step 5 Preview
Input price—$1/M tokens
Output price—$2.7/M tokens
Context window—1.0M tokens
WeightsOpen—
Free API (OpenRouter)NoNo
ReleasedDec 2025Oct 2026
At 10M a month–not listed$10.00$10.00
1M10M100M1B10M tokens

Input tokens at list price. No caching, no batch discount.

Where to run it2 hosts
Devstral 2 25121 host
HostInOutContextUptime
  • Mistral$0.40 in·$2.00 out·262k·100% up
Step 5 Preview1 host
HostInOutContextUptime
  • SStepFunfp8$1.00 in·$2.70 out·1M·99.2% up

Per million tokens. Prices and uptime via OpenRouter, checked 10 Oct 2026.

Common questions

What is the difference between Devstral 2 2512 and Step 5 Preview?

Devstral 2 2512 is developed by Mistral AI while Step 5 Preview is developed by StepFun. You can compare their actual outputs across 34 challenges on Rival to see how they differ in practice.

Which is better, Devstral 2 2512 or Step 5 Preview?

It depends on your use case. Devstral 2 2512 and Step 5 Preview each have strengths in different areas. Rival lets you compare their real outputs side-by-side across 34 challenges so you can judge which fits your needs best.

How can I compare Devstral 2 2512 and Step 5 Preview on Rival?

This page shows a side-by-side comparison of Devstral 2 2512 and Step 5 Preview across shared challenges. You can vote on which model produced the better output in a blind duel. Browsing and voting are free. No account is needed to look; signing in only saves your votes and likes.

More comparisons

Against the newest arrivals

  • Devstral 2 2512 vs Claude Haiku 5.5Landed Oct 2026
  • Step 5 Preview vs Ling 3.1 FlashLanded Oct 2026
  • Devstral 2 2512 vs Mistral Large 4Landed Oct 2026
  • Step 5 Preview vs GPT-6.1 SolLanded Sep 2026
  • Devstral 2 2512 vs Claude Sonnet 5.5Landed Sep 2026
  • Step 5 Preview vs Solar Mini 4Landed Sep 2026
  • Devstral 2 2512 vs Qwen3.8 Max PrimeLanded Sep 2026
  • Step 5 Preview vs GLM 5.3 PrimeLanded Sep 2026

Same lab, same size, long tail

  • Devstral 2 2512 vs Mistral Small 4Same lab
  • Devstral 2 2512 vs Ling 3.1 FlashSame size
  • Step 5 Preview vs Mistral Large 4Same size
  • Step 5 Preview vs Hy3Same size
  • Step 5 Preview vs Qwen3 Coder FlashNew provider
  • Step 5 Preview vs Qwen3 Coder NextNew provider
  • Step 5 Preview vs Qwen3 Coder PlusNew provider
  • Step 5 Preview vs Qwen3 MaxSame size

Model pages

  • Devstral 2 251235 outputs, specs and price
  • Step 5 Preview58 outputs, specs and price
Workshop · 150 skills

Build your own version.

Editable skills and working examples for your agent.

Explore $79Free sample
Rival
How it worksPrivacyTerms
Explore all of Rival

Explore

  • Compare models
  • All models
  • Image comparison
  • Audio comparison
  • Image generation
  • Best AI for...
  • Arena
  • API pricing
  • Challenges

Discover

  • SubjectiveBench
  • Default Index
  • Research
  • Research downloads
  • Rival Kits
  • Find your AI taste
  • UI Glow-Up
  • VoiceLock
  • Cost Cutter
  • Agent skills
  • Benchmarks vs Vibes
  • Brief
  • Jailbreak
  • Model updates
  • Provider status
  • AI creators

Connect

  • Methodology
  • Advertise
  • Partnerships
  • Privacy policy
  • Terms
  • RSS feed