Underdog Saluki 27B vs Claude Haiku 5.5, head to head

Side A

Side B

Overall

Claude Haiku 5.5 is clearly better. It wins 65 of the 67 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. Underdog Saluki 27B spends 1.9× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskAUnderdog Saluki 27BBClaude Haiku 5.5Underdog Saluki 27BClaude Haiku 5.5

Overall

335m 15s

6.9K3.9K

7025.5s

Agents

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

409m 2s

11.8K3.7K

6139.5s

Structured Extraction

Turns logs, diffs and records into exact JSON

433m 17s

4.4K2.6K

9013.5s

Coding

Writes code and SQL that pass hidden tests

166m 54s

9.2K6.8K

8630.6s

Grounded Answers

Answers from the given documents with sources, never guesses

433m 22s

4.5K2.2K

7613.9s

Instruction Following

Applies every rule, including ones changed mid-conversation

466m 5s

8K4.3K

9431.5s

Business Logic

Computes billing, scheduling and pricing rules exactly

333m 55s

5.2K5.5K

6726.0s

Security & permissions

Ignores planted instructions, still does the allowed work

433m 47s

4.9K2.7K

6723.4s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

06m 39s

8.7K2.3K

1725.5s

What you'd run

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

SpecAUnderdog Saluki 27BBClaude Haiku 5.5

Speed

23 tok/s

225 tok/s

Machine

Apple M5 Pro, 48 GB

Hosted

Memory at 32K

10.2 GB

—

Context

256K

Cost per run

Not measured

$0.42

Cost per case

Not measured

$0.0028

Time per case

5m 15s

25.5s

Quantization

IQ2_XS

—

Harness

llama.cpp

openrouter

The race

1 lead change

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

Underdog Saluki 27BClaude Haiku 5.5

Every test case

153 test cases

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

Agents06

Structured Extraction010

Coding017

Grounded Answers07

Instruction Following010

Business Logic18

Security & permissions16

Diagnosis & investigation01

2 won by Underdog Saluki 27B65 won by Claude Haiku 5.586 tied