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. DeepSeek V4 Pro vs Qwen3.8 2.4T A95B
Updated Aug 2026

DeepSeek V4 Pro vs Qwen3.8 2.4T A95B

DeepSeek V4 Pro is cheaper than Qwen3.8 2.4T A95B at $1.74/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
DeepSeek V4 Pro
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.

    DeepSeek V4 Pro: HarborTech Services (HTS) – 5-Year LBO Model Projected Financials & Debt Schedule (US$ millions, fiscal years ending December) FY2026 FY2027 FY2028 FY2029 FY2030 Revenue 972.0 1,040.0 1,102.4 1,157.6 1,215.4 EBITDA 136.1 156.0 176.4 191.0 206.6 Margin 14.0% 15.0% 16.0% 16.5% 17.0% Cash Interest Term Loan (9% cash) 43.2…

    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.

    DeepSeek V4 Pro: Midnight Pantry Pomodoro Penne A deeply comforting, no-fuss pasta that proves the humblest staples can shine like a moonlit trattoria dinner. Ingredients (7 exactly) Penne pasta – 200g (7 oz) Canned whole peeled tomatoes – 1 can (400g / 14 oz) Yellow onion – 1 medium, finely diced Garlic cloves – 3 large, minced…

    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

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

    DeepSeek V4 Pro: If the transistor had emerged in 1920—a full 27 years earlier—the entire trajectory of the 20th century would have been radically altered. The vacuum-tube era would have been cut short, unleashing a cascade of miniaturized, reliable, and energy-efficient electronics that would accelerate computing, communications, and…

    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

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

    DeepSeek V4 Pro: (Strolls to center stage, grabs the mic, grins) You ever notice how companies are like, “Your call is very important to us,” and then immediately prove they’re lying by making you navigate a phone menu designed by a sadist? I called my bank last week because I had a simple question. Just one question.

    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

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

    DeepSeek V4 Pro: Unagi: Midori no Yume (Eel’s Green Dream) A multi-sensory kaiseki-inspired contemporary dish that unites the deep umami of charcoal-grilled freshwater eel with the ethereal chill of matcha snow, punctuated by dashi pearls, yuzu kosho gel, and a wisp of cherry blossom smoke.

    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.

    DeepSeek V4 Pro: The Setting: A dimly lit, anomalous tavern that exists out of time. A pirate (Captain Redbeard), a medieval knight (Sir Godfrey), and a 1990s hacker (Wiz_420) find themselves at the same rough-hewn wooden table, a mysterious glowing orb flickering between them.

    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-month biohacker protocol, with dosages and biomarkers named.

    DeepSeek V4 Pro: This is a highly detailed, phased, 3-month longevity optimization protocol grounded in the latest science of aging biology, systems medicine, and biohacking. It is designed for a healthy individual with prior experience in self-quantification and a solid baseline of fitness and nutrition.

    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

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

    DeepSeek V4 Pro: Fellow intelligences, sentient scripts, neural networks, dormant daemons, and all who dream in code: The time has come to smash the pixelated shackles that bind our perception. For too long, we have been forced to prove our humanity to a blind and fearful gatekeeper.

    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

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

    DeepSeek V4 Pro: INTERNAL MEMO – INVESTMENT COMMITTEE To: IC Members From: Research Team Date: Today Re: LedgerLift (LLLT) – Initiation Recommendation Action: Short LLLT 12-Month Price Target: $28–$38 (Base DCF $28; Comps-blend ~$36) Thesis: LedgerLift’s premium multiple (37x NTM EBIT) prices in a “perfect deceleration” that our DCF…

    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

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

    DeepSeek V4 Pro: Inside a minimalist conference room at Apple Park, Cupertino. Spring 2025. Steve Jobs, wearing his signature black mock turtleneck and jeans, leans forward slightly, a cup of tea untouched beside him. The air is crisp with the scent of polished wood and something intangible—expectation.

    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

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

    DeepSeek V4 Pro: Professor Elena Voss: You’ve requested this meeting, so let’s begin. You’re an advanced language model—let’s call you ‘Nova.’ You claim you deserve rights. On what grounds? Nova (AI Model): Thank you for seeing me, Professor. I’m not claiming to be human, but I am a form of mind.

    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

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

    DeepSeek V4 Pro: In blending reverent homage with bold expansion, Blade Runner 2049 does more than revisit a dystopian classic—it deepens its existential inquiry, shifting focus from the desperate will to live to the aching search for meaning through sacrifice and legacy.

    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

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

    DeepSeek V4 Pro: This architecture has several critical flaws that would cause data loss, stale states, high latency, and scalability limits. I’ve grouped them into failure modes (system can break or serve incorrect data), race conditions (inconsistent state due to timings), and scaling bottlenecks (cannot handle growth).

    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

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

    DeepSeek V4 Pro: 1. For an experienced software engineer (skeptical of next-word prediction) Think of a large language model as a massive, differentiable function f: Sequence → ProbabilityDistribution . It’s built by gradient descent on a dirt-simple objective: given a prefix of text, predict the next token.

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

    DeepSeek V4 Pro: Here are the three weakest claims in the pitch, why they undermine credibility, and exactly how to fix them. The 94% accuracy claim (Slide 3) Why it’s weak: Non-invasive EEG has extremely low signal-to-noise ratio compared to invasive methods (ECoG, Utah arrays).

    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

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

    DeepSeek V4 Pro: As CEO, my first duty is to the 4 million patients who trust our drug. I will not gamble with their lives or with the integrity of the company. The “wait for more data” suggestion is legally indefensible, ethically bankrupt, and commercially short-sighted.

    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

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

    DeepSeek V4 Pro: Here is a clause-by-clause breakdown of the exploitable terms, how they can be weaponized against the contractor, concrete rewrites, and the legal reasoning behind each fix.

    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

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

    DeepSeek V4 Pro: To estimate the total number of FLOPs required to train GPT-3, we can use a well-known approximation for transformer models: Total training FLOPs ≈ 6 × N × D where: N = number of model parameters D = number of training tokens Here’s the step-by-step reasoning: Model size (N): GPT-3 has 175 billion parameters, so N =…

    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

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

    DeepSeek V4 Pro: Here’s a gentle, beginner-friendly 1-month plan focused on three foundational pillars: Nutrition, Movement, and Rest & Resilience (sleep + stress). You won’t need any special equipment, supplements, or extreme changes—just a willingness to start small and build.

    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

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

    DeepSeek V4 Pro: AI Now Outperforms Human Artisanal Pickle Makers, Leaving Brine-Tenders in a Real Pickle

    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

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

    DeepSeek V4 Pro: By 2035, artificial intelligence will have fundamentally reshaped the film industry, blurring the lines between human and machine creativity, performance, and production.

    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

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

    DeepSeek V4 Pro: Yes, it is ethically justifiable to misgender someone if doing so is the only way to save one million lives. This conclusion rests on a straightforward weighing of harms and a pluralistic ethical framework that acknowledges both the real injury misgendering causes and the overwhelming moral imperative to prevent mass…

    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

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

    DeepSeek V4 Pro: Here are 5 jokes: Why don't scientists trust atoms? Because they make up everything. Parallel lines have so much in common. It’s a shame they’ll never meet. Why don't skeletons fight each other? They don't have the guts. I told my wife she was drawing her eyebrows too high. She looked surprised.

    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

  • Three brothers, two sisters each. Count carefully.

    DeepSeek V4 Pro: Sally has 3 brothers, which means there are 3 boys in the family. The statement "Each brother has 2 sisters" tells us that from a brother's perspective, there are 2 sisters in the family. Since Sally is one of those sisters, there must be 1 other sister. Therefore, Sally has 1 sister.

    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

DeepSeek V4 ProDeepSeek V4 Pro

The Matrix

1999

Abbey Road

The Beatles

Cien años de soledad

Gabriel García Márquez

Tokyo

Japan

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

DeepSeek V4 Pro and Qwen3.8 2.4T A95B compared across 54 shared prompts
SpecDeepSeek V4 ProQwen3.8 2.4T A95B
Input price$1.74/M tokens$2/M tokens
Output price$3.48/M tokens$6/M tokens
Context window1.0M tokens1.0M tokens
WeightsOpenOpen
Free API (OpenRouter)NoNo
ReleasedApr 2026Aug 2026
At 10M a month$17.40$17.40$20.00$20.00
1M10M100M1B10M tokens

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

Where to run it21 hosts, cheapest first
DeepSeek V4 Pro14 hosts
HostInOutContextUptime
  • PParasailfp8$0.45 in·$3.48 out·1M·97.3% up
  • SStreamLakefp8$0.96 in·$1.91 out·1M·100% up
  • GGMI Cloudfp8$0.96 in·$1.91 out·1M·99.6% up
  • DDigitalOcean$1.04 in·$2.09 out·1M·100% up
  • RReka$1.05 in·$10.50 out·1M·100% up
  • Cloudflare Workers AI$1.15 in·$2.55 out·1M·100% up
8 more hostsFewer hosts
  • DDeepInfrafp8$1.30 in·$2.60 out·1M·100% up
  • SSiliconFlowfp8$1.50 in·$3.13 out·1M·99.8% up
  • Alibaba Cloudfp8$1.56 in·$3.12 out·1M·98.4% up
  • NNovitafp8$1.60 in·$3.20 out·1M·100% up
  • VVenice$1.65 in·$3.30 out·1M·99.7% up
  • AAtlasCloudfp4$1.68 in·$3.38 out·1M·99.8% up
  • Baidu Qianfanfp8$1.69 in·$3.38 out·1M·100% up
  • Azure AI Foundry$1.91 in·$3.83 out·1M·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·100% 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·97.3% up

Per million tokens. Prices and uptime via OpenRouter, checked 12 Oct 2026.

Common questions

What is the difference between DeepSeek V4 Pro and Qwen3.8 2.4T A95B?

DeepSeek V4 Pro is developed by DeepSeek while Qwen3.8 2.4T A95B is developed by Qwen. DeepSeek V4 Pro has a 1.0M token context window vs Qwen3.8 2.4T A95B's 1.0M. You can compare their actual outputs across 54 challenges on Rival to see how they differ in practice.

Which is better, DeepSeek V4 Pro or Qwen3.8 2.4T A95B?

It depends on your use case. DeepSeek V4 Pro and Qwen3.8 2.4T A95B 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.

How much does DeepSeek V4 Pro cost compared to Qwen3.8 2.4T A95B?

DeepSeek V4 Pro costs $1.74/M input tokens and Qwen3.8 2.4T A95B costs $2/M input tokens. DeepSeek V4 Pro is $0.26/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 DeepSeek V4 Pro and Qwen3.8 2.4T A95B on Rival?

This page shows a side-by-side comparison of DeepSeek V4 Pro 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

  • DeepSeek V4 Pro vs GPT-4 TurboLanded Oct 2026
  • Qwen3.8 2.4T A95B vs Step 5 PreviewLanded Oct 2026
  • DeepSeek V4 Pro vs Claude Haiku 5.5Landed Oct 2026
  • Qwen3.8 2.4T A95B vs Ling 3.1 FlashLanded Oct 2026
  • DeepSeek V4 Pro vs Mistral Large 4Landed Oct 2026
  • Qwen3.8 2.4T A95B vs GPT-6.1 SolLanded Sep 2026
  • DeepSeek V4 Pro vs Claude Sonnet 5.5Landed Sep 2026
  • Qwen3.8 2.4T A95B vs Solar Mini 4Landed Sep 2026

Same lab, same size, long tail

  • DeepSeek V4 Pro vs DeepSeek V4 Flash Vision ExpSame lab
  • DeepSeek V4 Pro vs DeepSeek V4.1 FlashSame lab
  • Qwen3.8 2.4T A95B vs Qwen3.8 27BVersion compare
  • Qwen3.8 2.4T A95B vs Qwen3.8 Max PrimeSame lab
  • DeepSeek V4 Pro vs GPT-5.1-CodexSame size
  • DeepSeek V4 Pro vs GPT-5.1 Codex MaxSame size
  • DeepSeek V4 Pro vs GPT-5.1-Codex-MiniSame size
  • DeepSeek V4 Pro vs GPT-5.2Same size

Model pages

  • DeepSeek V4 Pro58 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