Mellum2.1 Thinking vs Underdog Saluki 27B, head to head

Side A

Side B

Overall

Underdog Saluki 27B is clearly better. It wins 53 of the 56 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. Both models spend about the same on thinking per test case.

Thinking tokensAnswer tokens

TaskAMellum2.1 ThinkingBUnderdog Saluki 27BMellum2.1 ThinkingUnderdog Saluki 27B

Overall

91m 56s

10.3K6.9K

335m 15s

Agents

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

92m 3s

10.8K11.8K

409m 2s

Structured Extraction

Turns logs, diffs and records into exact JSON

51m 36s

8.3K4.4K

433m 17s

Coding

Writes code and SQL that pass hidden tests

52m 11s

11.3K9.2K

166m 54s

Grounded Answers

Answers from the given documents with sources, never guesses

131m 24s

7.3K4.5K

433m 22s

Instruction Following

Applies every rule, including ones changed mid-conversation

283m 10s

17.3K8K

466m 5s

Business Logic

Computes billing, scheduling and pricing rules exactly

01m 39s

9K5.2K

333m 55s

Security & permissions

Ignores planted instructions, still does the allowed work

81m 40s

9.1K4.9K

433m 47s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

01m 15s

6.8K8.7K

06m 39s

What you'd run

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

SpecAMellum2.1 ThinkingBUnderdog Saluki 27B

Speed

90 tok/s

23 tok/s

Machine

Apple M5 Pro, 48 GB

Memory at 32K

12.9 GB

10.2 GB

Context

128K

256K

Cost per run

$2.46 est.

Not measured

Cost per case

$0.016 est.

Not measured

Time per case

1m 56s

5m 15s

Quantization

Q8_0

IQ2_XS

Harness

llama.cpp

The race

No lead changes

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

Mellum2.1 ThinkingUnderdog Saluki 27B

Every test case

153 test cases

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

Agents19

Structured Extraction08

Coding17

Grounded Answers07

Instruction Following06

Business Logic07

Security & permissions19

Diagnosis & investigation00

3 won by Mellum2.1 Thinking53 won by Underdog Saluki 27B97 tied