LFM2.5 24B A2B vs Mistral Large 4, head to head

Side A

Side B

Overall

Mistral Large 4 is clearly better. It wins 44 of the 48 test cases where they differ.

Learn to build evals like this in the AI Engineering Academy

Task by task

A score turns orange only when that model wins clearly more of the task's test cases. Under each score, the average time per test case on the machine it ran on. LFM2.5 24B A2B spends 1.9× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskALFM2.5 24B A2BBMistral Large 4LFM2.5 24B A2BMistral Large 4

Overall

71m 53s

9.2K6.2K

281m 56s

Agents

Runs multi-step jobs through tools, within a written policy

123m 49s

18.5K795

1126.1s

Structured Extraction

Turns logs, diffs and records into exact JSON

1054.9s

4.4K288

153.78s

Coding

Writes code and SQL that pass hidden tests

11m 15s

6.1K9.3K

332m 48s

Grounded Answers

Answers from the given documents with sources, never guesses

557.8s

4.7K8.1K

382m 45s

Instruction Following

Applies every rule, including ones changed mid-conversation

162m 44s

13.4K12.9K

693m 47s

Business Logic

Computes billing, scheduling and pricing rules exactly

51m 2s

5K8.8K

242m 40s

Security & permissions

Ignores planted instructions, still does the allowed work

101m 52s

9K3.7K

331m 9s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

04m 12s

19.9K4.4K

01m 19s

What you'd run

The build behind each score. Orange marks the faster, smaller, cheaper or longer-context side.

SpecALFM2.5 24B A2BBMistral Large 4

Speed

82 tok/s

56 tok/s

Machine

Apple M5 Pro, 48 GB

Hosted

Memory at 32K

25.0 GB

—

Context

128K

512K

Cost per run

Not measured

$2.67

Cost per case

Not measured

$0.017

Time per case

1m 53s

1m 56s

Quantization

Q8_0

—

Harness

llama.cpp

openrouter

The race

3 lead changes

Each model's running score as the test cases go by. The dashed line is a perfect run.

LFM2.5 24B A2BMistral Large 4

Every test case

153 test cases

Each square is one test case, colored by the model that scored higher.

Agents22

Structured Extraction23

Coding08

Grounded Answers07

Instruction Following014

Business Logic04

Security & permissions06

Diagnosis & investigation00

4 won by LFM2.5 24B A2B44 won by Mistral Large 4105 tied