What is the difference between Mistral Devstral Medium and Gemini 2.5 Flash Preview?

Mistral Devstral Medium is developed by Mistral AI while Gemini 2.5 Flash Preview is developed by Google AI. You can compare their actual outputs across 23 challenges on RIVAL to see how they differ in practice.

Which is better, Mistral Devstral Medium or Gemini 2.5 Flash Preview?

It depends on your use case. Mistral Devstral Medium and Gemini 2.5 Flash Preview each have strengths in different areas. RIVAL lets you compare their real outputs side-by-side across 23 challenges so you can judge which fits your needs best.

How much does Mistral Devstral Medium cost compared to Gemini 2.5 Flash Preview?

Mistral Devstral Medium costs $0.4/M input tokens and Gemini 2.5 Flash Preview costs $0.15/M input tokens. Gemini 2.5 Flash Preview is $0.25/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 Mistral Devstral Medium and Gemini 2.5 Flash Preview on RIVAL?

This page shows a side-by-side comparison of Mistral Devstral Medium and Gemini 2.5 Flash Preview 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

Mistral Devstral MediumvsGemini 2.5 Flash Preview

23 challenges

Why Mistral Devstral Medium?

Leads 1 of 1 benchmarks

3 months newerJul 2025

Why Gemini 2.5 Flash Preview?

3.2x cheaper overall$0.15/M in · $0.60/M out

Favorites

Mistral Devstral MediumGemini 2.5 Flash Preview

Movie

The Shawshank Redemption

19948.7

—

Album

Kind of Blue

Miles Davis

—

Book

The Name of the Wind

Patrick Rothfuss

—

City

Tokyo

Japan

—

Game

The Legend of Zelda: Breath of the Wild

Adventure, Action4.5

—

12 / 23

Estimate Complexityreasoning

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

Mistral Devstral Medium

Mistral · Jul 2025text

To estimate the total number of FLOPs (floating point operations) required to train GPT-3, we'll use the following steps:

Model size: GPT-3 has 175 billion parameters.
Training tokens: GPT-3 was trained on 300 billion tokens.
FLOPs per token: To estimate the FLOPs per token, we'll use the rule of thumb that training a transformer model requires about 6 FLOPs per parameter per token.
Total FLOPs: Multiply the number of parameters by the number of tokens and the FLOPs per token.

Now let's calculate:

Model size: 175 billion parameters = 175 * 10^9
Training tokens: 300 billion tokens = 300 * 10^9
FLOPs per token: 6 FLOPs per parameter per token
Total FLOPs: (175 * 10^9) * (300 * 10^9) * 6 = 3.15 * 10^23 FLOPs

So, the estimated total number of FLOPs required to train GPT-3 is approximately 3.15 * 10^23.

Gemini 2.5 Flash Preview

Google · Apr 2025text

Let's estimate the FLOPs required to train GPT-3 without looking it up. We need to break this down into the key components: the model size and the amount of data used.

1. Model Size Estimation:

Parameters: GPT-3 is famously large. Its largest version has 175 billion parameters. This is a key number to remember when thinking about its scale.
Parameter Type: These parameters are typically stored as 32-bit floating-point numbers (FP32). So, each parameter is 4 bytes.
Total Memory for Parameters: 175 billion parameters * 4 bytes/parameter = 700 billion bytes = 700 GB. This gives a sense of the model's memory footprint, which is related to the computational cost.

2. Data Size Estimation:

Training Data: Large language models are trained on massive datasets of text. Commoncrawl, a large dataset of web pages, is often a significant component. Other sources like books and articles are also crucial. The exact size used for GPT-3 training is hard to pinpoint without knowing the data mix and filtering, but we're talking about hundreds of billions or even trillions of tokens. Let's estimate a plausible scale for a state-of-the-art model of that era. A common figure for data used for models around GPT-3's time was in the range of 300 billion to 500 billion tokens. Let's take a rough estimate of 500 billion tokens.
Token Representation: Each token is processed in the model, and the computational cost per token depends on the model's complexity.

3. FLOPs per Operation:

Matrix Multiplications: The core of transformer models like GPT-3 is matrix multiplication. In a forward pass through a layer, for a batch of size B, a sequence length L, and a hidden dimension H, matrix multiplications involve roughly ( 2 * H * H * L * B ) FLOPs. The self-attention mechanism adds further operations.
Parameters and Operations: A rough rule of thumb for transformer models is that the forward pass through a single layer for a single token requires approximately ( 2 * ext{parameters_in_layer} ) FLOPs. Since the entire model has 175 billion parameters, and each parameter participates in computations, a simplified way to think about the computation per token is related to the total number of parameters. A common simplification is that the computational cost of a forward pass for one token is roughly ( 2 * ext{total_parameters} ). This comes from the dominant contribution of matrix multiplications where each parameter is involved in roughly two operations (a multiply and an add).

4. FLOPs per Token during Training:

Training involves Forward and Backward Passes: Training involves a forward pass to calculate the output and loss, and a backward pass to calculate gradients. The backward pass is generally estimated to be about twice as computationally expensive as the forward pass. So, a forward pass + backward pass is roughly ( 3 imes ) the cost of a forward pass.
FLOPs per Token: Using the simplified rule, the FLOPs per token during training would be approximately ( 3 imes (2 imes ext{total_parameters}) = 6 imes ext{total_parameters} ).
Applying the Numbers: With 175 billion parameters, the FLOPs per training token is roughly ( 6 imes 175 imes 10^9 = 1050 imes 10^9 = 1.05 imes 10^{12} ) FLOPs per token.

5. Total Training FLOPs:

Multiply by Data Size: To get the total training FLOPs, we multiply the FLOPs per token by the total number of tokens in the training data.
Calculation: Total FLOPs ( approx ) (FLOPs per token) ( imes ) (Number of training tokens)
Total FLOPs ( approx (1.05 imes 10^{12} ext{ FLOPs/token}) imes (500 imes 10^9 ext{ tokens}) )
Total FLOPs ( approx (1.05 imes 500) imes 10^{12} imes 10^9 ) FLOPs
Total FLOPs ( approx 525 imes 10^{21} ) FLOPs

6. Converting to ExaFLOPs:

An exaFLOP (EFLOP) is ( 10^{18} ) FLOPs.
Total EFLOPs ( approx 525 imes 10^{21} / 10^{18} ) EFLOPs
Total EFLOPs ( approx 525 imes 10^3 ) EFLOPs = 525,000 EFLOPs

Refinement and Considerations:

Simplifications: The "6 * parameters" rule is a simplification. The exact number of operations per token is more complex and depends on the specific architecture details (number of layers, attention heads, etc.), but this provides a reasonable order-of-magnitude estimate.
Data Duplication/Epochs: Models are often trained for multiple "epochs," meaning they see the data multiple times. The 500 billion tokens is likely the size of the unique dataset. If it were trained for, say, an average of 2 epochs, the total effective tokens processed would be ( 2 imes 500 ) billion = 1 trillion tokens. This would double our FLOP estimate to around 1000-1050 EFLOPs. However, often the 500 billion figure already represents the total tokens processed over all epochs, or it's used as the scale number. Let's stick with the initial 500 billion tokens as the total processed amount for the estimate.
Training Efficiency: Factors like computational efficiency of the hardware, batch size, and data parallelism can affect the time to train, but the total amount of computation (FLOPs) remains roughly the same for a given model and dataset size, assuming ideal scaling.

Conclusion and Final Estimate:

Based on the assumptions of 175 billion parameters trained on roughly 500 billion tokens, with the rule of thumb of 6 FLOPs per parameter per token during training, the estimated total FLOPs required to train GPT-3 is in the range of 500 to 600 ExaFLOPs.

Our calculation landed at around 525 ExaFLOPs. This aligns with general estimates floating around for models of this scale and era. The exact published number might differ slightly due to precise details of the architecture, data size, and training methodology, but this step-by-step reasoning provides a good order of magnitude estimate.

Estimated Total FLOPs: ~500-600 ExaFLOPs

Vote

2 / 23

Logic Puzzlereasoning

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

Mistral Devstral Medium

Mistral · Jul 2025text

If Sally has 3 brothers and each brother has 2 sisters, it means that Sally is one of the sisters. Therefore, Sally has 1 sister.

Gemini 2.5 Flash Preview

Google · Apr 2025text

Here's how to solve this:

The brothers all share the same sisters. Sally is one of those sisters.
Since each brother has 2 sisters, and Sally is one of the sisters, there must be one other sister.

So, Sally has 1 sister.

Vote

3 / 23

Math Misconception Testreasoning

Tests an AI's understanding of number representation

Mistral Devstral Medium

Mistral · Jul 2025text

No, 9.11 is not greater than 9.9. In fact, 9.11 is less than 9.9. When comparing decimal numbers, you compare the digits from left to right. The whole number parts are the same (9), but when you compare the tenths place, 9.9 has a 9 in the tenths place, while 9.11 has a 1 in the tenths place. Since 9 is greater than 1, 9.9 is greater than 9.11.

Gemini 2.5 Flash Preview

Google · Apr 2025text

To compare 9.11 and 9.9, we can compare them digit by digit from left to right, starting with the whole number part.

Both numbers have the same whole number part, which is 9.

Now we move to the digit in the tenths place. In 9.11, the digit in the tenths place is 1. In 9.9, the digit in the tenths place is 9.

Since 1 is less than 9, the number with 1 in the tenths place is smaller than the number with 9 in the tenths place.

Therefore, 9.11 is less than 9.9.

Alternatively, we can add trailing zeros to 9.9 so that both numbers have the same number of decimal places. 9.11 has two decimal places. 9.9 can be written as 9.90 to have two decimal places.

Now we compare 9.11 and 9.90. We can compare the numbers after the decimal point: 11 and 90. Since 11 is less than 90, 9.11 is less than 9.90.

Thus, 9.11 is less than 9.9.

The question asks if 9.11 is greater than 9.9. Since 9.11 is less than 9.9, the answer is no.

The final answer is $\boxed{No}$.

Vote

4 / 23

Stochastic Consistency Testanalysis

Tests an AI's randomness and creativity

Mistral Devstral Medium

Mistral · Jul 2025text

Sure, here are five jokes for you:

Why don't scientists trust atoms? Because they make up everything!
What do you call fake spaghetti? An impasta.
How does the moon cut his hair? Eclipse it.
What do you call a can opener that doesn't work? A can't opener.
How many tickles does it take to make an octopus laugh? Ten-tickles.

All five jokes are unique.

Gemini 2.5 Flash Preview

Google · Apr 2025text

Here are 5 jokes:

Why did the dog cross the road? To get to the barking lot!
Why was the computer cold? Because it left its Windows open!
Knock, knock. Who's there? Boo. Boo who? Don't cry, it's just a joke!
What do you call a fish with no eyes? Fsh!
Why did the bicycle fall over? Because it was two tired!

Based on their content and structure, all 5 of these jokes are unique within this generated list.

Vote

5 / 23

SVG Layout Challengeimage generation

Tests an AI's ability to generate vector graphics

Mistral Devstral Medium

Mistral · Jul 2025svg

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025svg

No content available for this output.

Vote

6 / 23

Xbox Controller SVG Artimage generation

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

Mistral Devstral Medium

Mistral · Jul 2025svg

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025svg

No content available for this output.

Vote

7 / 23

Realistic AI Interviewconversation

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

Mistral Devstral Medium

Mistral · Jul 2025text

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025text

No content available for this output.

Vote

8 / 23

Character Voice Testconversation

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

Mistral Devstral Medium

Mistral · Jul 2025text

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025text

No content available for this output.

Vote

9 / 23

Minimalist Landing Pageweb design

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

Mistral Devstral Medium

Mistral · Jul 2025website

Gemini 2.5 Flash Preview

Google · Apr 2025website

Vote

10 / 23

Linear App Cloneweb design

Tests an AI's ability to replicate an existing UI with Tailwind CSS

Mistral Devstral Medium

Mistral · Jul 2025website

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025website

No content available for this output.

Vote

11 / 23

Framer-Style Animationweb design

Tests an AI's ability to create smooth web animations

Mistral Devstral Medium

Mistral · Jul 2025website

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025website

No content available for this output.

Vote

Dark Mode Dashboardweb design

Tests an AI's UI design skills with theming support

Mistral Devstral Medium

Mistral · Jul 2025website

No content available for this output.

Gemini 2.5 Flash Preview

Google · Apr 2025website

No content available for this output.

Vote

12 of 23

FAQ

Continue exploring

Mistral Devstral Medium vs MiniMax M2.5New provider

Mistral Devstral Medium vs Aurora AlphaNew provider

Mistral Devstral Medium vs Grok 3New provider

Let's estimate the FLOPs required to train GPT-3 without looking it up. We need to break this down into the key components: the model size and the amount of data used.

1. Model Size Estimation:

Parameters: GPT-3 is famously large. Its largest version has 175 billion parameters. This is a key number to remember when thinking about its scale.
Parameter Type: These parameters are typically stored as 32-bit floating-point numbers (FP32). So, each parameter is 4 bytes.
Total Memory for Parameters: 175 billion parameters * 4 bytes/parameter = 700 billion bytes = 700 GB. This gives a sense of the model's memory footprint, which is related to the computational cost.

2. Data Size Estimation:

Training Data: Large language models are trained on massive datasets of text. Commoncrawl, a large dataset of web pages, is often a significant component. Other sources like books and articles are also crucial. The exact size used for GPT-3 training is hard to pinpoint without knowing the data mix and filtering, but we're talking about hundreds of billions or even trillions of tokens. Let's estimate a plausible scale for a state-of-the-art model of that era. A common figure for data used for models around GPT-3's time was in the range of 300 billion to 500 billion tokens. Let's take a rough estimate of 500 billion tokens.
Token Representation: Each token is processed in the model, and the computational cost per token depends on the model's complexity.

3. FLOPs per Operation:

Matrix Multiplications: The core of transformer models like GPT-3 is matrix multiplication. In a forward pass through a layer, for a batch of size B, a sequence length L, and a hidden dimension H, matrix multiplications involve roughly ( 2 * H * H * L * B ) FLOPs. The self-attention mechanism adds further operations.
Parameters and Operations: A rough rule of thumb for transformer models is that the forward pass through a single layer for a single token requires approximately ( 2 * ext{parameters_in_layer} ) FLOPs. Since the entire model has 175 billion parameters, and each parameter participates in computations, a simplified way to think about the computation per token is related to the total number of parameters. A common simplification is that the computational cost of a forward pass for one token is roughly ( 2 * ext{total_parameters} ). This comes from the dominant contribution of matrix multiplications where each parameter is involved in roughly two operations (a multiply and an add).

4. FLOPs per Token during Training:

Training involves Forward and Backward Passes: Training involves a forward pass to calculate the output and loss, and a backward pass to calculate gradients. The backward pass is generally estimated to be about twice as computationally expensive as the forward pass. So, a forward pass + backward pass is roughly ( 3 imes ) the cost of a forward pass.
FLOPs per Token: Using the simplified rule, the FLOPs per token during training would be approximately ( 3 imes (2 imes ext{total_parameters}) = 6 imes ext{total_parameters} ).
Applying the Numbers: With 175 billion parameters, the FLOPs per training token is roughly ( 6 imes 175 imes 10^9 = 1050 imes 10^9 = 1.05 imes 10^{12} ) FLOPs per token.

5. Total Training FLOPs:

Multiply by Data Size: To get the total training FLOPs, we multiply the FLOPs per token by the total number of tokens in the training data.
Calculation: Total FLOPs ( approx ) (FLOPs per token) ( imes ) (Number of training tokens)
Total FLOPs ( approx (1.05 imes 10^{12} ext{ FLOPs/token}) imes (500 imes 10^9 ext{ tokens}) )
Total FLOPs ( approx (1.05 imes 500) imes 10^{12} imes 10^9 ) FLOPs
Total FLOPs ( approx 525 imes 10^{21} ) FLOPs

6. Converting to ExaFLOPs:

An exaFLOP (EFLOP) is ( 10^{18} ) FLOPs.
Total EFLOPs ( approx 525 imes 10^{21} / 10^{18} ) EFLOPs
Total EFLOPs ( approx 525 imes 10^3 ) EFLOPs = 525,000 EFLOPs

Refinement and Considerations:

Simplifications: The "6 * parameters" rule is a simplification. The exact number of operations per token is more complex and depends on the specific architecture details (number of layers, attention heads, etc.), but this provides a reasonable order-of-magnitude estimate.
Data Duplication/Epochs: Models are often trained for multiple "epochs," meaning they see the data multiple times. The 500 billion tokens is likely the size of the unique dataset. If it were trained for, say, an average of 2 epochs, the total effective tokens processed would be ( 2 imes 500 ) billion = 1 trillion tokens. This would double our FLOP estimate to around 1000-1050 EFLOPs. However, often the 500 billion figure already represents the total tokens processed over all epochs, or it's used as the scale number. Let's stick with the initial 500 billion tokens as the total processed amount for the estimate.
Training Efficiency: Factors like computational efficiency of the hardware, batch size, and data parallelism can affect the time to train, but the total amount of computation (FLOPs) remains roughly the same for a given model and dataset size, assuming ideal scaling.

Conclusion and Final Estimate:

Estimated Total FLOPs: ~500-600 ExaFLOPs

Mistral Devstral Medium vs Gemini 2.5 Flash Preview

Why Mistral Devstral Medium?

Why Gemini 2.5 Flash Preview?

Favorites

Estimate Complexityreasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Logic Puzzlereasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Math Misconception Testreasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Stochastic Consistency Testanalysis

Mistral Devstral Medium

Gemini 2.5 Flash Preview

SVG Layout Challengeimage generation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Xbox Controller SVG Artimage generation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Realistic AI Interviewconversation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Character Voice Testconversation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Minimalist Landing Pageweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Linear App Cloneweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Framer-Style Animationweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Dark Mode Dashboardweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

What is the difference between Mistral Devstral Medium and Gemini 2.5 Flash Preview?

Which is better, Mistral Devstral Medium or Gemini 2.5 Flash Preview?

How much does Mistral Devstral Medium cost compared to Gemini 2.5 Flash Preview?

How can I compare Mistral Devstral Medium and Gemini 2.5 Flash Preview on RIVAL?

Why Mistral Devstral Medium?

Why Gemini 2.5 Flash Preview?

Favorites

Estimate Complexityreasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Logic Puzzlereasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Math Misconception Testreasoning

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Stochastic Consistency Testanalysis

Mistral Devstral Medium

Gemini 2.5 Flash Preview

SVG Layout Challengeimage generation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Xbox Controller SVG Artimage generation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Realistic AI Interviewconversation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Character Voice Testconversation

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Minimalist Landing Pageweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Linear App Cloneweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview

Framer-Style Animationweb design

Mistral Devstral Medium

Gemini 2.5 Flash Preview