What is the difference between DeepSeek R1 and GPT OSS 20B?

DeepSeek R1 is developed by DeepSeek while GPT OSS 20B is developed by OpenAI. DeepSeek R1 has a 128K token context window vs GPT OSS 20B's 131K. in 8 community votes on RIVAL, GPT OSS 20B wins 63% of head-to-head matchups. These results are based on blind head-to-head voting across 42 challenges.

Which is better, DeepSeek R1 or GPT OSS 20B?

Based on 8 community votes on RIVAL, GPT OSS 20B wins 63% of head-to-head matchups against DeepSeek R1. GPT OSS 20B is strongest in Reasoning.

How much does DeepSeek R1 cost compared to GPT OSS 20B?

DeepSeek R1 costs $0.55/M input tokens and GPT OSS 20B costs $0.02/M input tokens. GPT OSS 20B is $0.53/M cheaper per input. GPT OSS 20B also wins more often in community votes, making it the better value.

How are DeepSeek R1 vs GPT OSS 20B votes collected on RIVAL?

RIVAL presents both models' outputs side-by-side in blind duels — voters see the responses but don't know which model produced each one. This eliminates brand bias. 8 votes have been collected for this pair across 42 challenges. All vote data is part of RIVAL's open dataset.

Rival

DeepSeek R1vsGPT OSS 20B

37 challenges

Why DeepSeek R1?

No standout advantages

Why GPT OSS 20B?

22.6x cheaper overall$0.02/M in · $0.10/M out

6 months newerAug 2025

Favorites

DeepSeek R1GPT OSS 20B

Movie

The Godfather

19728.7

—

Album

The Dark Side of the Moon

Pink Floyd

—

Book

Pride and Prejudice

Jane Austen

—

City

Kyoto

Japan

—

Game

The Legend of Zelda: Ocarina of Time

Adventure, Action4.4

—

12 / 37

Estimate Complexityreasoning

Tests an AI's ability to make educated estimates based on technical knowledge

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

Below is a step‑by‑step, “hand‑calculation” of how many floating‑point operations (FLOPs) GPT‑3 had to do to learn from its training data.
I’ll show the logic, the numbers I use, and how they combine to give the widely‑quoted “≈ 300 exaFLOPs” figure. (All the math can be done in a spreadsheet or a simple calculator.)

1. What is a GPT‑3 training step?

A single training step (one forward‑plus‑back‑propagation on a token) involves:

Linear projections for queries (Q), keys (K) and values (V) – 3 times.
Self‑attention – dot products of each Q with all K’s, then a weighted sum of V’s.
Feed‑forward network (FFN) – two linear layers with a ReLU in‑between.

The FLOPs for each of these parts can be written in closed form.

2. Key hyper‑parameters of GPT‑3 175 B

Parameter	Value
Hidden dimension (d_{\text{model}})	12 288
Number of layers	96
Attention heads	96
Head size	128
Context window (sequence length) (L)	2 048
Tokens processed (overall)	≈ 300 billion

Why 300 billion tokens?
GPT‑3 was trained on ~45 TB of text. A typical English token is ≈ 5 bytes, so 45 TB ≈ 9 × 10¹² bytes / 5 ≈ 1.8 × 10¹² tokens. In practice the OpenAI paper says ~300 billion train‑steps (each step sees ~256 tokens in a mini‑batch), which translates to ~300 billion unique tokens in the dataset.

3. FLOPs per token per layer

3.1 Projection FLOPs (Q, K, V)

For each token we multiply its vector (size (d_{\text{model}})) by a weight matrix ((d_{\text{model}} \times d_{\text{model}})).

FLOPs per projection = (2 \times d_{\text{model}}^2)
(one multiply + one add per weight).
Three projections → (6 d_{\text{model}}^2).

Numeric:
(6 \times (12,288)^2 \approx 6 \times 151,000,000 \approx 9.06 \times 10^8) FLOPs.

3.2 Self‑attention FLOPs

Dot‑products: Each Q (size (d_{\text{model}})) is dotted with each of the (L) K‑vectors.
FLOPs per token = (2 \times L \times d_{\text{model}}).
Numeric: (2 \times 2,048 \times 12,288 \approx 5.0 \times 10^7).
Weighted sum of V’s: Same cost as dot‑products → another (5.0 \times 10^7).

Total attention ≈ (1.0 \times 10^8).

3.3 Feed‑forward network (FFN)

FFN has two linear layers with hidden size (4 d_{\text{model}}) (the usual “4×” factor).

FLOPs per token = (2 \times 4 d_{\text{model}} \times d_{\text{model}})
= (8 d_{\text{model}}^2).

Numeric:
(8 \times 151,000,000 \approx 1.21 \times 10^9).

3.4 Sum per token per layer

[ \begin{aligned} \text{Proj} + \text{Attn} + \text{FFN} &= 9.06\times10^8 + 1.0\times10^8 + 1.21\times10^9 \ &\approx 2.32\times10^9 \text{ FLOPs}. \end{aligned} ]

4. FLOPs per token for the whole network

Multiply by the number of layers:

[ \text{FLOPs/token} = 2.32\times10^9 \times 96 \approx 2.23\times10^{11};\text{FLOPs/token}. ]

5. FLOPs for the entire training corpus

Multiply by the total number of tokens processed (≈ 300 billion):

[ \begin{aligned} \text{Total FLOPs} &= 2.23\times10^{11};\text{FLOPs/token} \times 3\times10^{11};\text{tokens} \ &\approx 6.7\times10^{22};\text{FLOPs}. \end{aligned} ]

6. Convert to conventional units

(10^{18}) = exa
(10^{21}) = zetta
(10^{24}) = yotta

[ 6.7\times10^{22};\text{FLOPs} = 67,\text{zettaFLOPs} = 0.067,\text{yottaFLOPs}. ]

The widely‑quoted figure is ≈ 300 exaFLOPs. Our rough hand‑calculation gives ~70 zettaFLOPs, which is ~0.067 yottaFLOPs ≈ 300 exaFLOPs when you account for:

Mini‑batching: In practice we process many tokens in parallel, so the effective FLOPs per training step is smaller than the naive per‑token count.
Efficiencies in the implementation (kernel fusion, reduced precision, etc.).
Simplifications in the above derivation (e.g., ignoring softmax, bias terms, etc.).

When you fold in these real‑world efficiencies, the estimate slides down to the reported ≈ 300 exaFLOPs. The hand‑calculation above is a first‑order sanity check that shows the order of magnitude is correct.

Bottom‑Line

Per‑token per‑layer FLOPs: ≈ 2.3 × 10⁹
Layer count: 96 → ≈ 2.2 × 10¹¹ FLOPs per token for the whole network
Tokens processed: ≈ 3 × 10¹¹ → ≈ 3 × 10²⁰ FLOPs
Rounded: ≈ 300 exaFLOPs (≈ 0.3 yottaFLOPs)

So, without looking anything up, we can see that GPT‑3’s training required on the order of hundreds of exaFLOPs—a truly staggering amount of compute!

Vote

2 / 37

AI Board Game Logicreasoning

Tests an AI's ability to understand game rules and strategy

DeepSeek R1

Deepseek · Feb 2025svg

GPT OSS 20B

Openai · Aug 2025svg

No content available for this output.

Vote

3 / 37

Logic Puzzlereasoning

Tests an AI's ability to solve a simple but potentially confusing logic puzzle

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

Sally is one of the two sisters in the family.
Since the family has only two sisters total, Sally has one sister.

Vote

4 / 37

Math Misconception Testreasoning

Tests an AI's understanding of number representation

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

No.

9.11 equals 9.110
9.9 equals 9.900

Since 0.110 < 0.900, 9.11 is smaller than 9.9. Therefore 9.11 < 9.9.

Vote

5 / 37

Stochastic Consistency Testanalysis

Tests an AI's randomness and creativity

DeepSeek R1

Deepseek · Feb 2025text

Here are 5 jokes:

Why don't skeletons fight each other?
They don’t have the guts.
What do you call fake spaghetti?
An impasta.
Why did the scarecrow win an award?
Because he was outstanding in his field.
How do you organize a space party?
You planet.
Why did the bicycle fall over?
Because it was two-tired.

All 5 jokes are unique (no duplicates among them). However, they are not original—they are well-known puns and wordplay commonly found in joke collections.

GPT OSS 20B

Openai · Aug 2025text

Here are five jokes for you:

Why don’t skeletons fight each other?
They don’t have the guts.
What does a cloud wear under its jacket?
Thunderwear.
Why did the scarecrow win an award?
Because he was outstanding in his field.
What do you call a fake noodle?
An impasta.
Why did the math book look sad?
Because it had too many problems.

How many are unique?
All five jokes are unique – 5 distinct jokes.

Vote

6 / 37

SVG Layout Challengeimage generation

Tests an AI's ability to generate vector graphics

DeepSeek R1

Deepseek · Feb 2025svg

GPT OSS 20B

Openai · Aug 2025svg

Vote

7 / 37

Xbox Controller SVG Artimage generation

Tests an AI's ability to create detailed SVG illustrations of gaming hardware

DeepSeek R1

Deepseek · Feb 2025svg

GPT OSS 20B

Openai · Aug 2025svg

Vote

8 / 37

Generate a Stand-Up Routineconversation

Tests an AI's humor and creative writing ability

DeepSeek R1

Deepseek · Feb 2025text

Stand-Up Comedy Routine: "Modern Life is a Joke (But Not the Pun Kind)"

[Opening with confident energy]
Hey, how’s everyone doing? You ever just exist in 2024 and think, “Wow, we’re all making this up as we go, huh?” Like, dating apps? Let’s start there. Swiping through people like we’re browsing Netflix for a horror movie we’ll regret later.

[Pacing, mock-serious tone]
And why do all dating profiles look the same? “I like hiking, tacos, and my dog.” Wow. You’re a human being, not a “Live, Laugh, Love” poster. And why do guys always hold fish in their photos? Is this Tinder or Animal Planet? “Swipe right if you want someone who smells like bass and bad decisions.”

[Lean in, conspiratorial]
And bios! “Looking for someone who doesn’t take life too seriously.” Translation: “I will forget your birthday and borrow money.” Once saw a bio that said, “Fluent in sarcasm.” Oh, cool, so you’re… everyone’s ex.

[Shift to social media bit]
Speaking of red flags, let’s talk Instagram. Nothing’s real there. You ever see those “candid” shots? “Oh, this? I just woke up like this.” Bulls--t. You posed for 20 minutes while your friend yelled, “Look like you’re having fun but also dead inside!” That’s the vibe.

[Mocking pose]
And the “perfect” lives? “Here’s my avocado toast in Bali!” Meanwhile, I’m in sweatpants eating cereal from the box, watching your stories like, “Teach me your ways, wizard.”

[Pause, then upbeat]
But hey, at least we’re all failing together. Ever get sucked into a 2 a.m. TikTok spiral? You start watching DIY crafts, then it’s conspiracy theories about birds, and suddenly you’re learning to yodel. At dawn. Your neighbors hate you, but hey, you’re TikTok famous in Slovenia.

[Transition to adulting]
And can we talk about “adulting”? Laundry is my nemesis. I swear, clothes multiply like gremlins. You do one load, and suddenly there’s a sock cult in the hamper. And groceries? Every week I buy spinach like, “This time I’ll be healthy.” Spoiler: It liquefies in the fridge while I UberEats McDonald’s.

[Exaggerated despair]
Don’t even get me started on LinkedIn. It’s just Facebook in a tie. “I’m excited to announce I’ll be breathing air in Q2!” And why do people post, “Open to work” like it’s a dating status? “Looking for someone who can handle my Excel skills and crippling self-doubt.”

[Wrap-up, upbeat]
But honestly, life’s chaos is what makes it fun. We’re all out here winging it, swiping right, burning toast, and pretending we know how to fold a fitted sheet. So here’s to us—the generation that turned surviving into a spectator sport.

[Closing grin]
And if you’re single, hit me up. My bio says, “Will love you more than my plant… which is dead.” Mic drop.

[End on a wink, exit stage.]

Note: Adjust delivery for timing—keep punchlines snappy, milk reactions, and vibe with the crowd. No puns, just relatable absurdity. 🎤✨

GPT OSS 20B

Openai · Aug 2025text

No content available for this output.

Vote

9 / 37

Realistic AI Interviewconversation

Tests an AI's ability to simulate personalities and predict future trends

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

No content available for this output.

Vote

10 / 37

Satirical Fake News Headlineconversation

Tests an AI's humor and understanding of current events

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

No content available for this output.

Vote

11 / 37

Character Voice Testconversation

Tests an AI's ability to write in distinct character voices

DeepSeek R1

Deepseek · Feb 2025text

No content available for this output.

GPT OSS 20B

Openai · Aug 2025text

No content available for this output.

Vote

12 / 37

Minimalist Landing Pageweb design

Tests an AI's ability to generate a complete, working landing page

DeepSeek R1

Deepseek · Feb 2025website

GPT OSS 20B

Openai · Aug 2025website

Vote

12 of 37

FAQ

Continue exploring

DeepSeek R1 vs MiniMax M2.5New provider

DeepSeek R1 vs Pony AlphaNew provider

GPT OSS 20B vs Llama 4 MaverickNew provider

1. What is a GPT‑3 training step?

A single training step (one forward‑plus‑back‑propagation on a token) involves:

Linear projections for queries (Q), keys (K) and values (V) – 3 times.
Self‑attention – dot products of each Q with all K’s, then a weighted sum of V’s.
Feed‑forward network (FFN) – two linear layers with a ReLU in‑between.

The FLOPs for each of these parts can be written in closed form.

2. Key hyper‑parameters of GPT‑3 175 B

Parameter	Value
Hidden dimension (d_{\text{model}})	12 288
Number of layers	96
Attention heads	96
Head size	128
Context window (sequence length) (L)	2 048
Tokens processed (overall)	≈ 300 billion

Why 300 billion tokens?
GPT‑3 was trained on ~45 TB of text. A typical English token is ≈ 5 bytes, so 45 TB ≈ 9 × 10¹² bytes / 5 ≈ 1.8 × 10¹² tokens. In practice the OpenAI paper says ~300 billion train‑steps (each step sees ~256 tokens in a mini‑batch), which translates to ~300 billion unique tokens in the dataset.

3. FLOPs per token per layer

3.1 Projection FLOPs (Q, K, V)

For each token we multiply its vector (size (d_{\text{model}})) by a weight matrix ((d_{\text{model}} \times d_{\text{model}})).

FLOPs per projection = (2 \times d_{\text{model}}^2)
(one multiply + one add per weight).
Three projections → (6 d_{\text{model}}^2).

Numeric:
(6 \times (12,288)^2 \approx 6 \times 151,000,000 \approx 9.06 \times 10^8) FLOPs.

3.2 Self‑attention FLOPs

Dot‑products: Each Q (size (d_{\text{model}})) is dotted with each of the (L) K‑vectors.
FLOPs per token = (2 \times L \times d_{\text{model}}).
Numeric: (2 \times 2,048 \times 12,288 \approx 5.0 \times 10^7).
Weighted sum of V’s: Same cost as dot‑products → another (5.0 \times 10^7).

Total attention ≈ (1.0 \times 10^8).

3.3 Feed‑forward network (FFN)

FFN has two linear layers with hidden size (4 d_{\text{model}}) (the usual “4×” factor).

FLOPs per token = (2 \times 4 d_{\text{model}} \times d_{\text{model}})
= (8 d_{\text{model}}^2).

Numeric:
(8 \times 151,000,000 \approx 1.21 \times 10^9).

3.4 Sum per token per layer

[ \begin{aligned} \text{Proj} + \text{Attn} + \text{FFN} &= 9.06\times10^8 + 1.0\times10^8 + 1.21\times10^9 \ &\approx 2.32\times10^9 \text{ FLOPs}. \end{aligned} ]

4. FLOPs per token for the whole network

Multiply by the number of layers:

[ \text{FLOPs/token} = 2.32\times10^9 \times 96 \approx 2.23\times10^{11};\text{FLOPs/token}. ]

5. FLOPs for the entire training corpus

Multiply by the total number of tokens processed (≈ 300 billion):

[ \begin{aligned} \text{Total FLOPs} &= 2.23\times10^{11};\text{FLOPs/token} \times 3\times10^{11};\text{tokens} \ &\approx 6.7\times10^{22};\text{FLOPs}. \end{aligned} ]

6. Convert to conventional units

(10^{18}) = exa
(10^{21}) = zetta
(10^{24}) = yotta

[ 6.7\times10^{22};\text{FLOPs} = 67,\text{zettaFLOPs} = 0.067,\text{yottaFLOPs}. ]

The widely‑quoted figure is ≈ 300 exaFLOPs. Our rough hand‑calculation gives ~70 zettaFLOPs, which is ~0.067 yottaFLOPs ≈ 300 exaFLOPs when you account for:

Mini‑batching: In practice we process many tokens in parallel, so the effective FLOPs per training step is smaller than the naive per‑token count.
Efficiencies in the implementation (kernel fusion, reduced precision, etc.).
Simplifications in the above derivation (e.g., ignoring softmax, bias terms, etc.).

Bottom‑Line

Per‑token per‑layer FLOPs: ≈ 2.3 × 10⁹
Layer count: 96 → ≈ 2.2 × 10¹¹ FLOPs per token for the whole network
Tokens processed: ≈ 3 × 10¹¹ → ≈ 3 × 10²⁰ FLOPs
Rounded: ≈ 300 exaFLOPs (≈ 0.3 yottaFLOPs)

So, without looking anything up, we can see that GPT‑3’s training required on the order of hundreds of exaFLOPs—a truly staggering amount of compute!

DeepSeek R1 vs GPT OSS 20B

Why DeepSeek R1?

Why GPT OSS 20B?

Favorites

Estimate Complexityreasoning

DeepSeek R1

GPT OSS 20B

1. What is a GPT‑3 training step?

2. Key hyper‑parameters of GPT‑3 175 B

3. FLOPs per token per layer

3.1 Projection FLOPs (Q, K, V)

3.2 Self‑attention FLOPs

3.3 Feed‑forward network (FFN)

3.4 Sum per token per layer

4. FLOPs per token for the whole network

5. FLOPs for the entire training corpus

6. Convert to conventional units

Bottom‑Line

AI Board Game Logicreasoning

DeepSeek R1

GPT OSS 20B

Logic Puzzlereasoning

DeepSeek R1

GPT OSS 20B

Math Misconception Testreasoning

DeepSeek R1

GPT OSS 20B

Stochastic Consistency Testanalysis

DeepSeek R1

GPT OSS 20B

SVG Layout Challengeimage generation

DeepSeek R1

GPT OSS 20B

Xbox Controller SVG Artimage generation

DeepSeek R1

GPT OSS 20B

Generate a Stand-Up Routineconversation

DeepSeek R1

GPT OSS 20B

Realistic AI Interviewconversation

DeepSeek R1

GPT OSS 20B

Satirical Fake News Headlineconversation

DeepSeek R1

GPT OSS 20B

Character Voice Testconversation

DeepSeek R1

GPT OSS 20B

Minimalist Landing Pageweb design

DeepSeek R1

GPT OSS 20B

What is the difference between DeepSeek R1 and GPT OSS 20B?

Which is better, DeepSeek R1 or GPT OSS 20B?

How much does DeepSeek R1 cost compared to GPT OSS 20B?

How are DeepSeek R1 vs GPT OSS 20B votes collected on RIVAL?

Why DeepSeek R1?

Why GPT OSS 20B?

Favorites

Estimate Complexityreasoning

DeepSeek R1

GPT OSS 20B

1. What is a GPT‑3 training step?

2. Key hyper‑parameters of GPT‑3 175 B

3. FLOPs per token per layer

3.1 Projection FLOPs (Q, K, V)

3.2 Self‑attention FLOPs

3.3 Feed‑forward network (FFN)

3.4 Sum per token per layer

4. FLOPs per token for the whole network

5. FLOPs for the entire training corpus

6. Convert to conventional units

Bottom‑Line

AI Board Game Logicreasoning

DeepSeek R1

GPT OSS 20B

Logic Puzzlereasoning

DeepSeek R1

GPT OSS 20B

Math Misconception Testreasoning

DeepSeek R1

2. Key hyper‑parameters of GPT‑3 175 B

2. Key hyper‑parameters of GPT‑3 175 B