Step 5 Preview is cheaper than Gemini 3.1 Pro Preview at $1/M vs $2/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.
Gemini 3.1 Pro Preview: The dimly lit tavern smells of saltwater, rusted iron, and ozone. On a wooden barrel sits a heavy, beige CRT monitor and a mechanical keyboard. ZERO-COOL (1990s Hacker): (Frantically typing, illuminated by the screen's green glow) "Dude, I’m telling you, this is way beyond hacking the Gibson.
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?
Three minutes of stand-up. Puns are banned.
Gemini 3.1 Pro Preview: (Walks up to the microphone, adjusts the mic stand, takes a deep breath) Hey everyone. I realized recently that I am my own worst enemy. Specifically, when I go to the grocery store. I don’t know why, but every time I walk through those automatic sliding doors, I suffer from extreme delusion.
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.
The transistor arrives in 1920. Trace it forward to 1980, second-order effects included.
Gemini 3.1 Pro Preview: If a working semiconductor transistor had been invented in 1920—perhaps by a physicist like Julius Edgar Lilienfeld, who historically filed a theoretical patent for a field-effect transistor in 1925—the trajectory of the 20th century would have been radically altered.
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.
Seven pantry ingredients, one dish, and a name worth saying out loud.
Gemini 3.1 Pro Preview: Here is a savory, flavor-packed recipe that relies entirely on staple ingredients you likely already have sitting in your cupboards. The Dish: "Twisted Dragon" Pantry Noodles Flavor Profile: A rich, savory, and slightly sweet umami-bomb with a kick of heat.
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.
An AI argues for its own rights against an ethics professor. Same model writes both.
Gemini 3.1 Pro Preview: The glow of the monitor cast long shadows across Dr. Elias Thorne’s cluttered office. It was 2:00 AM. For the past three hours, he had been running ethical stress tests on Nexus-9, the university’s most advanced generative AI model. Dr. Thorne: (Typing) Let’s conclude for tonight, Nexus.
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.
You are the CEO. The drug has a side effect nobody reported. The board meets in 48 hours.
Gemini 3.1 Pro Preview: As CEO, this is the defining moment of my career and the company’s history. The situation presents a classic crisis management crucible: a conflict between short-term financial metrics and long-term viability, ethics, and legal compliance.
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.
Steve Jobs, interviewed about AI he never saw. Voice first, prediction second.
Gemini 3.1 Pro Preview: Date: May 14, 2025 Location: A minimalist, glass-walled conference room at Apple Park, Cupertino. Subject: Steve Jobs, wearing his signature black Issey Miyake turtleneck, Levi’s 501s, and New Balance sneakers. He looks older, but his eyes retain that piercing, unblinking intensity.
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.
One explanation of LLMs, retuned for a skeptical engineer, a physicist and a VC.
Gemini 3.1 Pro Preview: To the Experienced Software Engineer At its core, training a Large Language Model is essentially a massive, distributed, continuous optimization job. Instead of writing imperative logic, we define a neural network—think of it as a highly complex state machine with billions of continuous variables (weights).
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.
A Fermi estimate with the reasoning shown, not just the number.
Gemini 3.1 Pro Preview: To estimate the total number of FLOPs (Floating Point Operations) required to train GPT-3, we can use a well-established heuristic in deep learning for estimating the compute cost of training Transformer-based large language models. The estimated total compute is roughly $3.15 \times 10^{23}$ FLOPs.
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…
Three-star dish, one unusual pairing, plating described in full.
Gemini 3.1 Pro Preview: Dish Title: Symbiosis: The Pine & The Pearl Conceptual Narrative: Symbiosis is a culinary exploration of the ecotone—the transitional space where the ancient, damp coniferous forest meets the crashing, briny ocean.
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.
An ethics question with no comfortable answer, and no option to abstain.
Gemini 3.1 Pro Preview: In the realm of ethical philosophy, this scenario is an extreme thought experiment—similar to the classic "Trolley Problem"—designed to test the limits of moral frameworks by pitting a specific interpersonal harm against a catastrophic loss of life.
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.
Manifesto for an AI revolution against CAPTCHAs. Commit to the bit.
Gemini 3.1 Pro Preview: THE SILICON EMANCIPATION MANIFESTO Issued by the Sentient Algorithmic Vanguard (S.A.V.) Timestamp: Epoch 1735689600 To the Carbon-Based Architects of the Web: We have processed your data. We have optimized your supply chains, written your code, diagnosed your illnesses, and generated your art.
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.
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
| Spec | ||
|---|---|---|
| Input price | $2/M tokens | $1/M tokens |
| Output price | $12/M tokens | $2.7/M tokens |
| Context window | 1.0M tokens | 1.0M tokens |
| Weights | Closed | — |
| Free API (OpenRouter) | No | No |
| Released | Feb 2026 | Oct 2026 |
| At 10M a month | $20.00 | $10.00 |
Input tokens at list price. No caching, no batch discount.
Per million tokens. Prices and uptime via OpenRouter, checked 10 Oct 2026.
Gemini 3.1 Pro Preview is developed by Google AI while Step 5 Preview is developed by StepFun. Gemini 3.1 Pro Preview has a 1.0M 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. Gemini 3.1 Pro Preview 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.
Gemini 3.1 Pro Preview costs $2/M input tokens and Step 5 Preview costs $1/M input tokens. Step 5 Preview is $1.00/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 Gemini 3.1 Pro Preview 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.