DeepSeek V4.1 Flash vs Underdog Saluki 27B, head to head

Side A

Side B

Overall

DeepSeek V4.1 Flash is clearly better. It wins 57 of the 63 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

TaskADeepSeek V4.1 FlashBUnderdog Saluki 27BDeepSeek V4.1 FlashUnderdog Saluki 27B

Overall

643m 35s

6K6.9K

335m 15s

Agents

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

464m 42s

6.1K11.8K

409m 2s

Structured Extraction

Turns logs, diffs and records into exact JSON

862m 8s

4.4K4.4K

433m 17s

Coding

Writes code and SQL that pass hidden tests

514m 55s

9.7K9.2K

166m 54s

Grounded Answers

Answers from the given documents with sources, never guesses

762m 17s

3.9K4.5K

433m 22s

Instruction Following

Applies every rule, including ones changed mid-conversation

894m 24s

7.5K8K

466m 5s

Business Logic

Computes billing, scheduling and pricing rules exactly

814m 1s

7K5.2K

333m 55s

Security & permissions

Ignores planted instructions, still does the allowed work

693m 3s

4.4K4.9K

433m 47s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

172m 8s

3.2K8.7K

06m 39s

What you'd run

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

SpecADeepSeek V4.1 FlashBUnderdog Saluki 27B

Speed

30 tok/s

23 tok/s

Machine

Hosted

Apple M5 Pro, 48 GB

Memory at 32K

—

10.2 GB

Context

256K

Cost per run

$0.39

Not measured

Cost per case

$0.0026

Not measured

Time per case

3m 35s

5m 15s

Quantization

FP8

IQ2_XS

Harness

openrouter

llama.cpp

The race

1 lead change

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

DeepSeek V4.1 FlashUnderdog Saluki 27B

Every test case

153 test cases

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

Agents53

Structured Extraction90

Coding102

Grounded Answers70

Instruction Following90

Business Logic100

Security & permissions61

Diagnosis & investigation10

57 won by DeepSeek V4.1 Flash6 won by Underdog Saluki 27B90 tied