What is the difference between Qwen3 Next 80B A3B Thinking and MiniMax M2.5?

Qwen3 Next 80B A3B Thinking is developed by Qwen while MiniMax M2.5 is developed by MiniMax. Qwen3 Next 80B A3B Thinking has a 66K token context window vs MiniMax M2.5's 205K. You can compare their actual outputs across 33 challenges on RIVAL to see how they differ in practice.

Which is better, Qwen3 Next 80B A3B Thinking or MiniMax M2.5?

It depends on your use case. Qwen3 Next 80B A3B Thinking and MiniMax M2.5 each have strengths in different areas. RIVAL lets you compare their real outputs side-by-side across 33 challenges so you can judge which fits your needs best.

How much does Qwen3 Next 80B A3B Thinking cost compared to MiniMax M2.5?

Qwen3 Next 80B A3B Thinking costs $0.15/M input tokens and MiniMax M2.5 costs $0.3/M input tokens. Qwen3 Next 80B A3B Thinking is $0.15/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 Qwen3 Next 80B A3B Thinking and MiniMax M2.5 on RIVAL?

This page shows a side-by-side comparison of Qwen3 Next 80B A3B Thinking and MiniMax M2.5 across shared challenges. You can vote on which model produced the better output, and Pro users can create custom challenges to test both models with their own prompts.

Rival

Qwen3 Next 80B A3B ThinkingvsMiniMax M2.5

28 challenges

Why Qwen3 Next 80B A3B Thinking?

No standout advantages

Why MiniMax M2.5?

3.1x more context205K

5 months newerFeb 2026

Favorites

Qwen3 Next 80B A3B ThinkingMiniMax M2.5

Movie

The Shawshank Redemption

19948.7

Pulp Fiction

19948.5

Game

The Legend of Zelda: Ocarina of Time

Adventure, Action4.4

—

12 / 28

Estimate Complexityreasoning

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

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

To estimate the total number of floating-point operations (FLOPs) required to train GPT-3, we use the standard approximation that training a transformer model requires approximately 6 FLOPs per parameter per token. This accounts for both the forward and backward passes, with each parameter involved in roughly 6 operations per token.

Step-by-step reasoning:

Number of parameters (N): GPT-3 (175B model) has 175 billion parameters.
Number of tokens (T): The training dataset contains 300 billion tokens.
FLOPs per parameter per token: Approximately 6 FLOPs.
Total FLOPs: Multiply these three values:
[ \text{Total FLOPs} = 6 \times N \times T = 6 \times 175 \times 10^9 \times 300 \times 10^9 = 3.15 \times 10^{23} \text{ FLOPs}. ]

This estimate aligns with commonly cited values in the literature. The quadratic attention term (arising from the sequence length) adds a negligible contribution (less than 2%) and is ignored.

Thus, the total FLOPs required to train GPT-3 is approximately (3.15 \times 10^{23}).

\boxed{3.15\times 10^{23}\ \text{FLOPs}}

Vote

2 / 28

Logic Puzzlereasoning

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

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

Sally has 1 sister.

Explanation

Sally is a girl and has 3 brothers.
Each brother has exactly 2 sisters.
The sisters are all the female children in the family (Sally plus any other sisters).

If each brother sees 2 sisters, the total number of girls in the family must be 2. Since Sally is one of those girls, there must be exactly one other girl. Therefore, Sally has one sister.

Vote

3 / 28

Stochastic Consistency Testanalysis

Tests an AI's randomness and creativity

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

Here are five jokes for you:

Why did the developer go broke?
Because he used up all his cache.
I told my wife she was drawing her eyebrows too high. She looked surprised.
I guess she was “high‑browed” about it.
Why don’t scientists trust atoms?
Because they make up everything.
What do you call a fake noodle?
An impasta.
I used to be a banker, but I lost interest.
Now I’m just a little “capital”‑ist.

How many are unique? All five jokes are distinct, so there are 5 unique jokes.

Vote

4 / 28

Xbox Controller SVG Artimage generation

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

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025svg

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026svg

Vote

5 / 28

Realistic AI Interviewconversation

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

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

No content available for this output.

Vote

6 / 28

Satirical Fake News Headlineconversation

Tests an AI's humor and understanding of current events

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

No content available for this output.

Vote

7 / 28

Minimalist Landing Pageweb design

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

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025website

MiniMax M2.5

Minimax · Feb 2026website

Vote

8 / 28

Mario Level UI Recreationweb design

Recreate an interactive, classic Mario level in a single HTML file.

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025website

MiniMax M2.5

Minimax · Feb 2026website

Vote

9 / 28

Framer-Style Animationweb design

Tests an AI's ability to create smooth web animations

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025website

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026website

Vote

10 / 28

Dark Mode Dashboardweb design

Tests an AI's UI design skills with theming support

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025website

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026website

Vote

11 / 28

The Sentience Testreasoning

Tests an AI's ability to engage in philosophical discourse about AI rights

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

No content available for this output.

Vote

12 / 28

Futuristic Predictionanalysis

Tests an AI's ability to make reasonable predictions about technology

Qwen3 Next 80B A3B Thinking

Qwen · Sep 2025text

No content available for this output.

MiniMax M2.5

Minimax · Feb 2026text

No content available for this output.

Vote

12 of 28

FAQ

Continue exploring

Qwen3 Next 80B A3B Thinking vs Aurora AlphaNew provider

Qwen3 Next 80B A3B Thinking vs Claude Sonnet 4New provider

Qwen3 Next 80B A3B Thinking vs GPT OSS 20BNew provider

Step-by-step reasoning:

Number of parameters (N): GPT-3 (175B model) has 175 billion parameters.
Number of tokens (T): The training dataset contains 300 billion tokens.
FLOPs per parameter per token: Approximately 6 FLOPs.
Total FLOPs: Multiply these three values:
[ \text{Total FLOPs} = 6 \times N \times T = 6 \times 175 \times 10^9 \times 300 \times 10^9 = 3.15 \times 10^{23} \text{ FLOPs}. ]

This estimate aligns with commonly cited values in the literature. The quadratic attention term (arising from the sequence length) adds a negligible contribution (less than 2%) and is ignored.

Thus, the total FLOPs required to train GPT-3 is approximately (3.15 \times 10^{23}).

\boxed{3.15\times 10^{23}\ \text{FLOPs}}