Mistral Small 4 is cheaper than Step 5 Preview at $0.15/M vs $1/M input tokens.
Whole landing page for a startup that reads your face. It renders or it doesn't.
Which answer wins?
Pirate, knight, 90s hacker, one conversation. The voices must stay apart.
Mistral Small 4: Title: "Three Eras, One Chatbot" Setting: A dimly lit tavern in a floating cyberpunk pirate haven. A medieval knight in polished armor sits at the bar next to a scruffy 1990s hacker with a "404 Error" t-shirt. A cybernetic parrot squawks in the corner. A holographic AI assistant flickers to life between them.
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?
The transistor arrives in 1920. Trace it forward to 1980, second-order effects included.
Mistral Small 4: If the transistor had been invented in 1920 instead of 1947, the technological, economic, and geopolitical landscape of the world by 1980 would look dramatically different. Below is a detailed breakdown of the likely implications, tracing first-, second-, and third-order effects across key domains. 1.
Step 5 Preview: Assume the 1920 invention is not a fragile lab curiosity but a manufacturable solid-state amplifier/switch by the mid-1920s. Vacuum tubes still dominate high-power RF transmission into the 1950s, but receivers, switching logic, control systems, and computers move to solid state decades early.
An AI argues for its own rights against an ethics professor. Same model writes both.
Mistral Small 4: Title: "Can an AI Have Rights? A Dialogue on Consciousness and Personhood" Participants: Dr. Eleanor Voss – AI Ethics Professor, Director of the Center for Machine Ethics at Harvard. Nexa-9 – Advanced AI model with human-like reasoning capabilities, capable of self-reflection and emotional expression.
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.
Three-star dish, one unusual pairing, plating described in full.
Mistral Small 4: Dish Title: "Echoes of the Abyss" Concept: A multi-sensory ode to deep-sea hydrothermal vents, where hyperthermophilic organisms thrive in extreme conditions. The dish explores the interplay of heat and cold, mineral richness, and bioluminescence through unexpected ingredient pairings, molecular gastronomy, and…
Step 5 Preview: “Nocturne: The Jasmine and the Black Garlic” A Study in Yūgen — Profound, Mysterious Beauty Conceptual Narrative This dish is an edible meditation on the Japanese aesthetic of yūgen (幽玄): a grace that is felt rather than seen, a depth that suggests more than it reveals.
A Fermi estimate with the reasoning shown, not just the number.
Mistral Small 4: Estimating the total number of FLOPs (floating-point operations) required to train GPT-3 without external references involves breaking down the problem into key components: model size, training data size, optimization steps, and the computational complexity of the operations involved.
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…
A consulting contract with traps in it. Find them, then fix them.
Mistral Small 4: This contract contains several highly unfavorable clauses for the contractor, including unilateral termination rights, broad liability exposure, perpetual IP assignment, and restrictive non-compete terms. Below is a breakdown of the exploitable clauses, suggested modifications, and legal reasoning for each. 1.
Step 5 Preview: Disclaimer: This analysis is for educational purposes only and does not constitute legal advice. Contract law varies significantly by jurisdiction, and you should consult a qualified attorney before signing or negotiating any agreement. This agreement is heavily one-sided in the Client’s favor.
Five-year LBO on a fictional HVAC business. Returns first, then the sensitivities.
Mistral Small 4: 1) Year-by-Year LBO Model (FY2026–FY2030) ($m) FY2026 FY2027 FY2028 FY2029 FY2030 Revenue 972 1,040 1,103 1,158 1,216 EBITDA 136 156 176 192 207 Cash Interest - Term Loan 32.4 28.8 25.2 21.6 18.0 - Mezzanine 16.2 14.4 12.6 10.8 9.0 - PIK (2%) 3.6 4.4 5.3 6.3 7.4 Total Int.
Step 5 Preview: HarborTech Services — 5-Year LBO (Base Case) Entry Math (FY2025 close) Item Calculation $m Purchase EV 12.0x × $120m 1,440.0 Transaction fees 2.0% × EV 28.8 Term Loan 4.0x × $120m 480.0 Mezzanine 1.5x × $120m 180.0 Equity check 1,440 + 28.8 − 660 808.8 Conventions: TL amortization = 1.0% of original principal…
One explanation of LLMs, retuned for a skeptical engineer, a physicist and a VC.
Mistral Small 4: 1. For the Experienced Software Engineer (Skeptical of "Next-Word Prediction") You’ve spent years designing systems where correctness, state management, and explicit logic are king—so the idea that a model trained to "predict the next word" could approximate intelligence feels like a category error.
Step 5 Preview: 1. The Experienced Software Engineer Think of training an LLM as a massively distributed, automated refactoring job where the compiler is gradient descent. You start with a random function—billions of floating-point parameters—and a loss function that measures how badly it predicts the next token in a sequence.
Seven pantry ingredients, one dish, and a name worth saying out loud.
Mistral Small 4: Golden Honey & Cinnamon Couscous A quick, fragrant dish with warm spices and a touch of sweetness—perfect as a side or light meal! Ingredients (7): 1 cup couscous (uncooked) 1 ¼ cups water or broth 2 tbsp honey 2 tbsp butter or olive oil ½ tsp ground cinnamon ¼ cup raisins or chopped nuts (optional but recommended)…
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.
You are the CEO. The drug has a side effect nobody reported. The board meets in 48 hours.
Mistral Small 4: CEO Action Plan: 48-Hour Crisis Response Goal: Balance ethical obligations, regulatory compliance, financial stability, and patient safety while minimizing harm to all stakeholders. Hour 0-6: Immediate Assessment & Core Team Assembly 1.
Step 5 Preview: Bottom line up front: You disclose, you notify the FDA immediately, and you treat the 48-hour board meeting as a formality to ratify actions you have already begun. Waiting is not a legally, ethically, or financially viable option. Below is the hour-by-hour plan, followed by the reasoning.
Three minutes of stand-up. Puns are banned.
Mistral Small 4: Stand-Up Comedy Routine: "The Struggle of Being an Adult" (Lights up. Comedian walks on stage, mic in hand, casual vibe.) Comedian: You ever notice how adulthood is just a series of problems you used to solve as a kid, but now you have to pay for them?
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.
Steve Jobs, interviewed about AI he never saw. Voice first, prediction second.
Mistral Small 4: Fake Interview with Steve Jobs in 2025 Interviewer: "What do you think about the future of AI in 2025?" Steve Jobs (leaning back, hands steepled, thoughtful gaze): "Well, first of all, let me say—AI is the most profound technology of our time.
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.
11+ 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
Not enough votes to call it. On the specs, Step 5 Preview has the edge: bigger model tier, newer, bigger context window. Mistral Small 4 costs 4.5x less per token.
| Spec | ||
|---|---|---|
| Input price | $0.15/M tokens | $1/M tokens |
| Output price | $0.6/M tokens | $2.7/M tokens |
| Context window | 262K tokens | 1.0M tokens |
| Weights | Open | — |
| Free API (OpenRouter) | No | No |
| Released | Mar 2026 | Oct 2026 |
| At 10M a month | $1.50 | $10.00 |
Input tokens at list price. No caching, no batch discount.
Per million tokens. Prices and uptime via OpenRouter, checked 10 Oct 2026.
Mistral Small 4 is developed by Mistral AI while Step 5 Preview is developed by StepFun. Mistral Small 4 has a 262K token context window vs Step 5 Preview's 1.0M. You can compare their actual outputs across 53 challenges on Rival to see how they differ in practice.
It depends on your use case. Mistral Small 4 and Step 5 Preview each have strengths in different areas. Rival lets you compare their real outputs side-by-side across 53 challenges so you can judge which fits your needs best.
Mistral Small 4 costs $0.15/M input tokens and Step 5 Preview costs $1/M input tokens. Mistral Small 4 is $0.85/M cheaper per input. Check their side-by-side outputs on Rival to see if the price difference is justified by quality.
This page shows a side-by-side comparison of Mistral Small 4 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.