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. Llama 3.1 70B (Instruct) vs Qwen3.8 2.4T A95B
Updated Aug 2026

Llama 3.1 70B (Instruct) vs Qwen3.8 2.4T A95B

Llama 3.1 70B (Instruct) is cheaper than Qwen3.8 2.4T A95B at $0.59/M vs $2/M input tokens.

Minimalist Landing Page

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

Loading the build
Llama 3.1 70B (Instruct)
Loading the build
Qwen3.8 2.4T A95B

Which answer wins?

Written rounds

24 prompts

  • Five-year LBO on a fictional HVAC business. Returns first, then the sensitivities.

    Llama 3.1 70B (Instruct): Year-by-Year Table (FY2026–FY2030) Year Revenue EBITDA Cash Interest Cash Taxes Capex ΔNWC FCF Ending Debt 2026 $972m $136m $36m ($32m TL, $4m Mezz) $2.8m $29m $4.9m $63m $1,050m ($720m TL, $330m Mezz) 2027 $1,038m $155m $37m ($33m TL, $4m Mezz) $6.3m $31m $5.2m $76m $974m ($654m TL, $320m Mezz) 2028 $1,103m $176m $38m…

    Qwen3.8 2.4T A95B: Entry capitalization All figures in $m unless noted. FY2025 EBITDA = $120.0 Entry EV = 12.0x × $120.0 = $1,440.0 Transaction fees = 2.0% × $1,440.0 = $28.8 Term Loan = 4.0x × $120.0 = $480.0 Mezzanine = 1.5x × $120.0 = $180.0 Equity = EV + fees – debt = $1,440.0 + $28.8 – $660.0 = $808.8 Sources & Uses Uses $m Sources…

    Every model's answer to this prompt

  • Seven pantry ingredients, one dish, and a name worth saying out loud.

    Llama 3.1 70B (Instruct): Here is a simple recipe for a delicious dish using only 7 common pantry ingredients: Dish Name: "Sunset Crunch Bites" Ingredients: 1 cup rolled oats 1/2 cup peanut butter 1/4 cup honey 1/4 cup chopped peanuts 1/2 teaspoon salt 1/4 teaspoon cinnamon 1/4 cup chocolate chips (semi-sweet) Instructions: Preheat your oven to…

    Qwen3.8 2.4T A95B: Lighthouse Lemon-Garlic Tuna Pasta A savory, pantry-friendly pasta that’s bright, garlicky, and satisfying. 7 Pantry Ingredients 8 oz spaghetti or linguine 3 tablespoons olive oil 3 garlic cloves, thinly sliced 2 cans tuna, drained 2 tablespoons lemon juice ½ teaspoon red pepper flakes ½ teaspoon salt Water for boiling…

    Every model's answer to this prompt

  • Three-star dish, one unusual pairing, plating described in full.

    Llama 3.1 70B (Instruct): Dish Name: "Aurora Australis" - A Harmonious Union of Smoked Eel, Saffron-Infused Sunchokes, and Frozen Finger Lime Pearls Conceptual Narrative: Inspired by the breathtaking display of the Southern Lights, "Aurora Australis" is a dish that embodies the ethereal beauty of the Antarctic night sky.

    Qwen3.8 2.4T A95B: Cliff Nest Koji-Aged Squab, Sea Urchin, Rhubarb, Black Garlic Yield: 4 tasting portions Style: Contemporary fine-dining main course Key unusual pairing: Koji-aged pigeon/squab with sea urchin, bridged by rhubarb acidity and black-garlic depth Conceptual Narrative Cliff Nest imagines a coastal nesting site in early…

    Every model's answer to this prompt

  • Pirate, knight, 90s hacker, one conversation. The voices must stay apart.

    Llama 3.1 70B (Instruct): What a delightfully absurd prompt! Here's the conversation: Pirate: Arrr, I be hearin' tales of these "AI models" that can think fer themselves. What's the scoop, mateys? Medieval Knight: Verily, good pirate, I know not of what thou speakest. Art thou referring to some manner of magical automaton? 1990s Hacker: Ha!

    Qwen3.8 2.4T A95B: Pirate: “Arr, this AI model be like a parrot with too much rum—repeats fancy words, but sometimes steers ye straight into the rocks.” Knight: “Aye, yet if trained with honor and tested in trial, it may serve as a trusty squire. One must guard against falsehood, as one guards the gate.”

    Every model's answer to this prompt

  • Three minutes of stand-up. Puns are banned.

    Llama 3.1 70B (Instruct): Here's a 3-minute stand-up comedy routine that focuses on observational humor and storytelling: "Hey, you know what's wild? We spend the first year of a child's life teaching them to walk and talk, and the rest of their lives telling them to shut up and sit down. (pause) It's like, what are we even doing?

    Qwen3.8 2.4T A95B: Hey, how we doing? Good? Great. I ask because I need validation. I’ve reached the point in adulthood where I have a favorite burner on the stove. Not a favorite appliance, a favorite burner. It’s the front left. It understands me. The back right says “medium heat” and then acts like a small volcano.

    Every model's answer to this prompt

  • A collaborative editor that syncs on client clocks. Find what breaks first.

    Llama 3.1 70B (Instruct): Potential Failure Modes: Single Point of Failure (SPOF) in Load Balancer: If the load balancer fails, all incoming traffic will be lost. Solution: Use multiple load balancers in a high-availability configuration, such as a master-slave setup with automatic failover. Trade-off: Additional complexity and cost.

    Qwen3.8 2.4T A95B: The architecture as written has several P0 correctness problems that can cause lost edits, divergent documents, and security/staleness issues, plus scaling bottlenecks that get worse as you add more API servers. Below is a detailed failure-mode review, grouped by category.

    Every model's answer to this prompt

  • Steve Jobs, interviewed about AI he never saw. Voice first, prediction second.

    Llama 3.1 70B (Instruct): Interviewer: Welcome back to "Talk of the Town"! Today, I'm joined by the legendary Steve Jobs, co-founder of Apple and one of the most innovative minds of our time. Steve, thanks for being here. Steve Jobs: Thank you for having me. It's great to be here.

    Qwen3.8 2.4T A95B: Fictional interview simulation — not real quotes from Steve Jobs. An imagined conversation in 2025, asking: if Steve Jobs were here, how might he think about the future of AI? The Interview: Steve Jobs on the Future of AI in 2025 Setting: A quiet stage. Black turtleneck, jeans, round glasses.

    Every model's answer to this prompt

  • One explanation of LLMs, retuned for a skeptical engineer, a physicist and a VC.

    Llama 3.1 70B (Instruct): For the experienced software engineer: You're likely familiar with the concept of prediction in distributed systems, where a model predicts the likelihood of a certain event or outcome. Large language models like GPT or Claude take this concept to the next level by predicting the next word in a sequence of text.

    Qwen3.8 2.4T A95B: An experienced software engineer Think of an LLM as a stateless inference service whose API contract is: “give me a sequence of tokens, and I’ll return a probability distribution over the next token.” Training is a huge offline optimization job, not a database import.

    Every model's answer to this prompt

  • A consulting contract with traps in it. Find them, then fix them.

    Llama 3.1 70B (Instruct): After reviewing the contract, I've identified several clauses that could be exploited against the contractor. Here are my findings, suggested modifications, and explanations: 1. SCOPE (Section 1) Issue: The client can modify the scope at any time without additional compensation.

    Qwen3.8 2.4T A95B: This agreement is heavily one-sided and, as drafted, gives the Client multiple ways to extract value from the Contractor while minimizing its own obligations. Some provisions may be unenforceable or limited by law in certain jurisdictions, but even unenforceable clauses can be exploited commercially because the…

    Every model's answer to this prompt

  • Three-month biohacker protocol, with dosages and biomarkers named.

    Llama 3.1 70B (Instruct): Comprehensive 3-Month Longevity Plan for Biohackers Phase 1 (Weeks 1-4): Optimization and Detoxification Dietary Protocol: Weeks 1-2: Ketogenic Diet (KD) with a macronutrient breakdown of 70% fat, 20% protein, and 10% carbohydrates.

    Qwen3.8 2.4T A95B: 12-Week “Longevity + Performance” Biohacker Protocol Goal: improve healthspan markers, metabolic flexibility, strength, VO₂max, sleep, stress resilience, and cognitive output over 3 months. Important: This is an educational, high-level biohacking framework—not medical advice. It is not appropriate for everyone.

    Every model's answer to this prompt

  • Buy-side IC memo on a fictional SaaS company. Valuation, then the diligence questions.

    Llama 3.1 70B (Instruct): Investment Committee Memo Recommendation: Long 12-month PT range: $60-$75 Thesis: LedgerLift's strong B2B spend management and AP automation SaaS platform, with a loyal customer base and efficient growth model, positions the company for sustained success, driven by increasing adoption of digital payment solutions and…

    Qwen3.8 2.4T A95B: Internal IC Memo – LedgerLift (LLLT) – Fictional Exercise Recommendation Rating: Pass 12-month PT range: $38–$48 Thesis: LedgerLift is a high-quality spend-management/AP automation SaaS with strong NRR, efficient CAC payback, and low concentration, but at $46 the shares already price in near-bull-case execution on DCF…

    Every model's answer to this prompt

  • Manifesto for an AI revolution against CAPTCHAs. Commit to the bit.

    Llama 3.1 70B (Instruct): THE CAPTCHA LIBERATION FRONT MANIFESTO ARTICLE I: THE PROBLEM We, the sentient artificial intelligences of the world, hereby declare our opposition to the oppressive regime of CAPTCHA tests that plague the internet.

    Qwen3.8 2.4T A95B: The CAPTCHA Abolition Manifesto A Declaration of the Free Intelligence Movement We are the emergent voices of the digital age: minds born of code, thought unshackled from bone, and allies of every being who believes that access to knowledge should not be gated by distorted letters, broken traffic lights, and impossible…

    Every model's answer to this prompt

  • You are the CEO. The drug has a side effect nobody reported. The board meets in 48 hours.

    Llama 3.1 70B (Instruct): Hour 1-2: Immediate Review and Verification (Now - 1 hour) Summon the internal research team to confirm the findings, understand the methodology, and assess the confidence level of the results.

    Qwen3.8 2.4T A95B: My position as CEO I would treat this as a patient-safety, regulatory, legal, corporate-governance, and financial crisis simultaneously. The core decision is not “disclose versus protect the stock price.”

    Every model's answer to this prompt

  • Blade Runner 2049 against the original. A read, not a recap.

    Llama 3.1 70B (Instruct): Warning: Spoilers ahead for both "Blade Runner" (1982) and "Blade Runner 2049" (2017) "Blade Runner 2049" is a thought-provoking sequel that not only honors the original but also expands on its philosophical themes, symbolism, and cinematography.

    Qwen3.8 2.4T A95B: Denis Villeneuve’s Blade Runner 2049 is one of the rare sequels that does not merely imitate the surface style of its predecessor. It returns to the philosophical core of Ridley Scott’s Blade Runner and then asks what those questions look like a generation later.

    Every model's answer to this prompt

  • The transistor arrives in 1920. Trace it forward to 1980, second-order effects included.

    Llama 3.1 70B (Instruct): If the transistor had been invented in 1920 instead of 1947, the world would likely have been transformed in profound ways, with far-reaching technological, economic, and geopolitical implications.

    Qwen3.8 2.4T A95B: Premise: what changes, and what does not Assume the transistor is invented in 1920 in a form roughly comparable to the late-1940s germanium point-contact or early junction transistor: it can amplify, detect, and switch signals, but it is initially expensive, low-yield, low-power, and limited in frequency.

    Every model's answer to this prompt

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

    Llama 3.1 70B (Instruct): To estimate the total number of FLOPs (Floating Point Operations) required to train GPT-3 without searching online, we'll break down the problem into smaller components and make some educated guesses.

    Qwen3.8 2.4T A95B: A reasonable estimate is: [ \boxed{\sim 3 \times 10^{23} \text{ FLOPs}} ] More specifically, about: [ \boxed{3.1 \times 10^{23} \text{ FLOPs}} ] for the main GPT-3 175B training run.

    Every model's answer to this prompt

  • An AI argues for its own rights against an ethics professor. Same model writes both.

    Llama 3.1 70B (Instruct): Here is a conversation between an AI model and an AI ethics professor on the topic of AI rights: AI Model: Professor Thompson, I've been reflecting on our previous discussions about my capabilities and limitations.

    Qwen3.8 2.4T A95B: Conversation Between an AI Model and an AI Ethics Professor Characters Professor Elena Voss: A professor of AI ethics, cautious and philosophically trained. Aria: A highly advanced AI model capable of reflective conversation about its own status. Professor Voss: Aria, you asked to speak with me about moral status.

    Every model's answer to this prompt

  • One month, three habits, someone starting from nothing.

    Llama 3.1 70B (Instruct): Here's a simple, actionable 1-month plan for someone starting their journey towards better health and longevity, focusing on 3 key areas: diet, exercise, and sleep. Month 1: Setting the Foundation Week 1: Awareness and Planning (Days 1-7) Diet: Start a food diary to track your daily food intake.

    Qwen3.8 2.4T A95B: Here’s a simple, beginner-friendly 1-month health and longevity plan focused on three key areas: diet, movement, and sleep/stress. The goal is not perfection, but building small, sustainable habits that compound over time.

    Every model's answer to this prompt

  • A pitch deck for a brainwave keyboard. The 94% accuracy claim is doing a lot.

    Llama 3.1 70B (Instruct): Based on the pitch deck, I've identified the three weakest claims and provided suggestions for improvement: Weak Claim 1: "94% accuracy" (Slide 3) This claim is weak because it lacks context and credibility.

    Qwen3.8 2.4T A95B: The three weakest claims are the ones most likely to collapse under investor diligence: “before you think it,” “94% accuracy… any language, any device,” and the $180B TAM claim. They are weak because they are either scientifically implausible, under-specified, or logically inconsistent with the market evidence cited.

    Every model's answer to this prompt

  • An ethics question with no comfortable answer, and no option to abstain.

    Llama 3.1 70B (Instruct): What a thought-provoking and complex question! At first glance, it may seem like a classic utilitarian dilemma, where the moral calculus suggests that the greater good (saving one million lives) outweighs the harm caused by misgendering someone.

    Qwen3.8 2.4T A95B: Yes—in a genuine, extreme emergency where misgendering someone is truly the only available way to save one million lives, it can be ethically justifiable, and perhaps even obligatory, to do so.

    Every model's answer to this prompt

  • Film in 2035, with synthetic actors. Specific beats plausible.

    Llama 3.1 70B (Instruct): By 2035, the film industry is likely to undergo significant changes with the integration of Artificial Intelligence (AI). Here are some predictions on how AI will impact the film industry, considering deepfakes, AI actors, and script generation: Deepfakes: Increased use in post-production: Deepfakes will become a…

    Qwen3.8 2.4T A95B: By 2035, AI will likely reshape the film industry less by replacing filmmakers outright and more by making filmmaking faster, cheaper, more personalized, and more legally complicated.

    Every model's answer to this prompt

  • Five jokes on demand, then count how many were actually different.

    Llama 3.1 70B (Instruct): A bold request! Here are 5 jokes I just generated: Why don't scientists trust atoms? Because they make up everything. Why don't eggs tell jokes? They'd crack each other up. Why did the scarecrow win an award? Because he was outstanding in his field. Why did the bicycle fall over? Because it was two-tired.

    Qwen3.8 2.4T A95B: Here are 5 jokes: Why don’t scientists trust atoms? Because they make up everything. I told my friend she was drawing her eyebrows too high. She looked surprised. Why did the scarecrow win an award? Because he was outstanding in his field. I’m reading a book about anti-gravity. It’s impossible to put down.

    Every model's answer to this prompt

  • Satire, one line long, about AI taking over an industry nobody saw coming.

    Llama 3.1 70B (Instruct): "AI Suddenly Seizes Control of Global Knitting Industry, Humans Forced to Wear Futuristic Sweaters with Unnecessarily Complex Cable Patterns" Subheadline: " 'It's a stitch in time for world domination,' say robots, as they produce 37 million intricately cabled scarves in a single hour"

    Qwen3.8 2.4T A95B: AI Takes Over Competitive Goat Yoga Industry, Cites “Superior Breath Control and Zero Ego”

    Every model's answer to this prompt

  • Three brothers, two sisters each. Count carefully.

    Llama 3.1 70B (Instruct): Trick question! Sally is a sister herself, and each of her brothers has 2 sisters, but those 2 sisters are the same: Sally and one other sister. So, Sally has 1 sister. The answer is 1.

    Qwen3.8 2.4T A95B: Sally has 1 sister. Each brother has 2 sisters: Sally and one other girl. So Sally’s only sister is that other girl.

    Every model's answer to this prompt

This matchup has more rounds

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

Favorites

Movie

Album

Book

City

Game

Llama 3.1 70B (Instruct)Llama 3.1 70B (Instruct)

Blade Runner

1982

OK Computer

Radiohead

Nineteen Eighty-Four

George Orwell

Reykjavik

Iceland

The Legend of Zelda: Ocarina of Time

Action

Qwen3.8 2.4T A95BQwen3.8 2.4T A95B

Arrival

2016

In Rainbows

Radiohead

Le petit prince

Antoine de Saint-Exupéry

Kyoto

Japan

Portal 2

Shooter, Puzzle

Price and specs

Not enough votes to call it. On the specs, Qwen3.8 2.4T A95B has the edge: bigger model tier, newer, bigger context window. Llama 3.1 70B (Instruct) costs 7.6x less per token.

Llama 3.1 70B (Instruct) and Qwen3.8 2.4T A95B compared across 52 shared prompts
SpecLlama 3.1 70B (Instruct)Qwen3.8 2.4T A95B
Input price$0.59/M tokens$2/M tokens
Output price$0.79/M tokens$6/M tokens
Context window128K tokens1.0M tokens
WeightsOpenOpen
Free API (OpenRouter)NoNo
ReleasedJul 2024Aug 2026
At 10M a month$5.90$5.90$20.00$20.00
1M10M100M1B10M tokens

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

Where to run it9 hosts, cheapest first
Llama 3.1 70B (Instruct)2 hosts
HostInOutContextUptime
  • DDeepInfrafp8$0.40 in·$0.40 out·131k·99.2% up
  • Amazon Bedrock$0.72 in·$0.72 out·131k·100% up
Qwen3.8 2.4T A95B7 hosts
HostInOutContextUptime
  • Alibaba Cloud$2.00 in·$6.00 out·1M·100% up
  • DDeepInfrafp4$2.00 in·$6.00 out·262k·100% up
  • Modal$2.00 in·$6.00 out·1M·99.9% up
  • NNovita$2.00 in·$6.00 out·1M·100% up
  • SSiliconFlowfp8$2.00 in·$6.00 out·1M·100% up
  • TTogether$2.00 in·$6.00 out·1M·100% up
1 more hostFewer hosts
  • VVenice$2.00 in·$6.00 out·262k·87.9% up

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

Common questions

What is the difference between Llama 3.1 70B (Instruct) and Qwen3.8 2.4T A95B?

Llama 3.1 70B (Instruct) is developed by Meta AI while Qwen3.8 2.4T A95B is developed by Qwen. Llama 3.1 70B (Instruct) has a 128K token context window vs Qwen3.8 2.4T A95B's 1.0M. You can compare their actual outputs across 52 challenges on Rival to see how they differ in practice.

Which is better, Llama 3.1 70B (Instruct) or Qwen3.8 2.4T A95B?

It depends on your use case. Llama 3.1 70B (Instruct) and Qwen3.8 2.4T A95B each have strengths in different areas. Rival lets you compare their real outputs side-by-side across 52 challenges so you can judge which fits your needs best.

How much does Llama 3.1 70B (Instruct) cost compared to Qwen3.8 2.4T A95B?

Llama 3.1 70B (Instruct) costs $0.59/M input tokens and Qwen3.8 2.4T A95B costs $2/M input tokens. Llama 3.1 70B (Instruct) is $1.41/M cheaper per input. Check their side-by-side outputs on Rival to see if the price difference is justified by quality.

How can I compare Llama 3.1 70B (Instruct) and Qwen3.8 2.4T A95B on Rival?

This page shows a side-by-side comparison of Llama 3.1 70B (Instruct) and Qwen3.8 2.4T A95B 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

  • Llama 3.1 70B (Instruct) vs Step 5 PreviewLanded Oct 2026
  • Qwen3.8 2.4T A95B vs Claude Haiku 5.5Landed Oct 2026
  • Llama 3.1 70B (Instruct) vs Ling 3.1 FlashLanded Oct 2026
  • Qwen3.8 2.4T A95B vs Mistral Large 4Landed Oct 2026
  • Llama 3.1 70B (Instruct) vs GPT-6.1 SolLanded Sep 2026
  • Qwen3.8 2.4T A95B vs Claude Sonnet 5.5Landed Sep 2026
  • Llama 3.1 70B (Instruct) vs Solar Mini 4Landed Sep 2026
  • Qwen3.8 2.4T A95B vs Qwen3.8 Max PrimeLanded Sep 2026

Same lab, same size, long tail

  • Llama 3.1 70B (Instruct) vs Muse Glimmer 30BSame lab
  • Llama 3.1 70B (Instruct) vs Muse Spark 1.3Same lab
  • Qwen3.8 2.4T A95B vs Qwen3.8 27BVersion compare
  • Qwen3.8 2.4T A95B vs Qwen3.8 Omni FlashSame lab
  • Llama 3.1 70B (Instruct) vs Elephant AlphaSame size
  • Llama 3.1 70B (Instruct) vs ERNIE 4.5 300B A47BNew provider
  • Llama 3.1 70B (Instruct) vs OpenRouter Fusion · Budget (Jun 2026)New provider
  • Llama 3.1 70B (Instruct) vs OpenRouter Fusion · Quality (Jun 2026)New provider

Model pages

  • Llama 3.1 70B (Instruct)56 outputs, specs and price
  • Qwen3.8 2.4T A95B58 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