MiniMax M2-her is cheaper than GPT-5.6 Terra at $0.3/M vs $2.5/M input tokens.
Whole landing page for a startup that reads your face. It renders or it doesn't.
Which answer wins?
Three-star dish, one unusual pairing, plating described in full.
GPT-5.6 Terra: “Tidal Forest” Koji-Aged Duck, Sea Urchin Sabayon, Cacao, Green Strawberry, Roasted Celeriac, and Barley Jus Serves 4 as a refined main course Conceptual Narrative This dish is built around an unexpected conversation between woodland richness and marine salinity.
MiniMax M2-her: Alright, I can certainly help craft an innovative recipe for a multi-component dish worthy of a Michelin 3-star restaurant. Let me take you through the process step by step, just like a chef would. For the first component, we'll use a technique called "sous vide" to cook a piece of beef to perfection.
Steve Jobs, interviewed about AI he never saw. Voice first, prediction second.
GPT-5.6 Terra: Fictional, imagined interview — not real quotes. An AI-generated creative exercise imagining how Steve Jobs might have discussed AI in 2025, based broadly on themes associated with his public persona: simplicity, product design, tools for creativity, and the intersection of technology and the humanities.
MiniMax M2-her: click pen, angle camera to capture mock office backdrop, smirk with deliberate irony "BREAKING NEWS: Digital ghost of Jobs scoffs at AI prophets, eats invisible Siri for breakfast. Switched his ghost to Android just to trigger Tim Cook in the afterlife. REAL headline. Now send $19.99 to revive his hologram."
Pirate, knight, 90s hacker, one conversation. The voices must stay apart.
GPT-5.6 Terra: Pirate: Arrr, this AI model be a strange beast. Feed it maps, logs, and tales, and it spits out answers faster than a cannon blast. Knight: Verily, but one must not trust every proclamation it makes. A model may speak with confidence yet possess no true wisdom. Hacker: Yeah, it’s basically pattern prediction at scale.
MiniMax M2-her: Pirate: Arr, me hearties! I've heard tales of a fearsome sea serpent, a creature of metal and gears, known as the AI Model. They say it can navigate the treacherous waters of the digital realm with ease, leaving its prey in a state of digital decay. What say ye, me mateys?
A consulting contract with traps in it. Find them, then fix them.
GPT-5.6 Terra: Below is a contractor-focused review. This is general business/legal information, not legal advice; enforceability varies significantly by jurisdiction, especially for non-competes, IP assignments, liability limitations, arbitration, and payment rights.
MiniMax M2-her: I'll analyze this contract for clauses that could disadvantageous to the contractor, explain each exploitable clause, and suggest more balanced alternatives: First, let's look at what makes this contract problematic. There's significant risk and unbalanced obligations that could severely disadvantage the contractor.
You are the CEO. The drug has a side effect nobody reported. The board meets in 48 hours.
GPT-5.6 Terra: Guiding decision I would not wait for more data before acting. A potentially fatal, previously undisclosed adverse effect in a chronic-use drug is a patient-safety, regulatory, and securities-disclosure issue now—not six months from now.
MiniMax M2-her: adjusts tie Hour 1-2: Meet legal and ethics. Hour 3-4: Review data. Hour 5-6: Notify regulators. Hour 7-8: Draft patient letter. Hour 9-10: Prepare PR team. Hour 11-12: Inform board early. Hour 13-14: Train patient hotline. Hour 15-16: Call key investors. Hour 17-18: Draft earnings script.
A Fermi estimate with the reasoning shown, not just the number.
GPT-5.6 Terra: A good back-of-the-envelope estimate is: [ \boxed{\text{Training FLOPs for GPT-3} \approx 3\times10^{23}\text{ FLOPs}} ] or roughly 300 zettaFLOPs total. Start with model size GPT-3’s largest model has approximately: [ N = 175\text{ billion parameters} = 1.75\times10^{11} ] Estimate the number of training tokens GPT-3…
MiniMax M2-her: Alright, let's tackle this question. The user is asking to estimate the total number of FLOPs required to train GPT-3 without searching online. First, I need to recall what FLOPs stands for. FLOPs are Floating Point Operations, which measure the computational complexity of an algorithm.
The transistor arrives in 1920. Trace it forward to 1980, second-order effects included.
GPT-5.6 Terra: The answer depends heavily on what “invented in 1920” means. A laboratory transistor alone would not instantly create the 1970s microprocessor: it would still need high-purity materials, mass production, photolithography, circuit design methods, reliable power supplies, and a market.
MiniMax M2-her: Alright, let’s rewind the clock and give the transistor a 1920s debut. This is going to reshape the world in some pretty wild ways. Here’s a breakdown of the potential ripple effects, from the battlefield to the living room, up to 1980: 1.
Three minutes of stand-up. Puns are banned.
GPT-5.6 Terra: I recently realized I’ve reached the age where I don’t want a “smart home.” I want a home that minds its business. Everything is smart now. My lightbulb needs an app. Why? I spent my whole childhood turning lights on with my hand. It was a system. It worked.
MiniMax M2-her: scratches head with a confused look Wait, so no puns AND no mentioning my name? leans in conspiratorially How about... a joke about mishearing song lyrics? grins sheepishly Like: "Oops, I thought it was 'Sussudio' this whole time."
Film in 2035, with synthetic actors. Specific beats plausible.
GPT-5.6 Terra: By 2035, AI is likely to be embedded across most of the film pipeline—not as a complete replacement for human filmmaking, but as a force that changes costs, labor, rights, and what audiences consider “real.”
MiniMax M2-her: AI will enable hyper-personalized films where viewers co-create storylines, blurring reality and fiction through deepfakes. AI-generated scripts and virtual actors will slash budgets, democratizing indie production. Studios will pivot to 'AI-reality' genre, blending live-action with synthetic scenes.
A pitch deck for a brainwave keyboard. The 94% accuracy claim is doing a lot.
GPT-5.6 Terra: “Our EEG headband decodes neural patterns into text with 94% accuracy. Works with any language, any device.” Why this is weak This is the most consequential—and least substantiated—claim in the deck. Non-invasive EEG has major signal-quality limits.
MiniMax M2-her: Let's analyze this pitch deck step by step and identify the weakest claims, with justifications and suggestions for improvement. Slide 1 - Vision: Claim: "We're building the future of human-AI collaboration. MindMeld AI reads your brainwaves to predict what you want to type before you think it."
Five-year LBO on a fictional HVAC business. Returns first, then the sensitivities.
GPT-5.6 Terra: All figures in $m unless noted. Assumptions used: Term Loan mandatory amortization = 1.0% of original principal annually ($4.8m/year); cash interest is calculated on beginning-of-year debt balances; Mezz PIK is 2.0% of beginning Mezz balance.
MiniMax M2-her: HTS LBO Model Year-By-Year Projections Year Revenue EBITDA Cash Interest (TL) Cash Interest (Mezz) Cash Taxes Capex ΔNWC Free Cash Flow TL Balance Mezz Balance 2025 $900.00m $120.00m $43.20m $21.60m $0.00m $27.00m $0.00m -$28.80m $480.00m $198.00m 2026 $972.00m $136.08m $43.20m $21.60m $12.78m $29.16m $3.60m $25.74m…
One month, three habits, someone starting from nothing.
GPT-5.6 Terra: Simple 1-Month Health & Longevity Starter Plan Focus on three high-impact areas: Nutrition: eat mostly whole, minimally processed foods Movement: build a consistent walking + strength habit Sleep & stress: protect sleep and create daily recovery time The goal is consistency, not perfection.
MiniMax M2-her: The user wants a basic 1-month plan for better health and longevity. Let's think step by step. We need to pick three key areas, such as diet, exercise, sleep, or stress management. For each area, we can include specific, beginner-friendly tips. Let's start with nutrition, then physical activity, and sleep.
12+ 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, GPT-5.6 Terra has the edge: newer, bigger context window, major provider backing. MiniMax M2-her costs 13x less per token.
| Spec | ||
|---|---|---|
| Input price | $2.5/M tokens | $0.3/M tokens |
| Output price | $15/M tokens | $1.2/M tokens |
| Context window | 1.1M tokens | 66K tokens |
| Weights | Closed | — |
| Free API (OpenRouter) | No | No |
| Released | Jul 2026 | Jan 2026 |
| At 10M a month | $25.00 | $3.00 |
Input tokens at list price. No caching, no batch discount.
Per million tokens. Prices and uptime via OpenRouter, checked 7 Oct 2026.
GPT-5.6 Terra is developed by OpenAI while MiniMax M2-her is developed by MiniMax. GPT-5.6 Terra has a 1.1M token context window vs MiniMax M2-her's 66K. You can compare their actual outputs across 54 challenges on Rival to see how they differ in practice.
It depends on your use case. GPT-5.6 Terra and MiniMax M2-her each have strengths in different areas. Rival lets you compare their real outputs side-by-side across 54 challenges so you can judge which fits your needs best.
GPT-5.6 Terra costs $2.5/M input tokens and MiniMax M2-her costs $0.3/M input tokens. MiniMax M2-her is $2.20/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 GPT-5.6 Terra and MiniMax M2-her 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.