DeepSeek V4.1 Flash vs Spark X2.6 4B, head to head

Side A

Side B

Overall

DeepSeek V4.1 Flash is clearly better. It wins 90 of the 92 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. Spark X2.6 4B spends 1.3× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskADeepSeek V4.1 FlashBSpark X2.6 4BDeepSeek V4.1 FlashSpark X2.6 4B

Overall

643m 35s

6K8.2K

151m 3s

Agents

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

464m 42s

6.1K11.5K

141m 32s

Structured Extraction

Turns logs, diffs and records into exact JSON

862m 8s

4.4K5.3K

1039.5s

Coding

Writes code and SQL that pass hidden tests

514m 55s

9.7K9.1K

41m 8s

Grounded Answers

Answers from the given documents with sources, never guesses

762m 17s

3.9K6.6K

1050.2s

Instruction Following

Applies every rule, including ones changed mid-conversation

894m 24s

7.5K11.9K

371m 29s

Business Logic

Computes billing, scheduling and pricing rules exactly

814m 1s

7K4.9K

1036.3s

Security & permissions

Ignores planted instructions, still does the allowed work

693m 3s

4.4K7K

3154.3s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

172m 8s

3.2K11.6K

01m 30s

What you'd run

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

SpecADeepSeek V4.1 FlashBSpark X2.6 4B

Speed

30 tok/s

136 tok/s

Machine

Hosted

RTX 5090, 32 GB

Memory at 32K

—

9.4 GB

Context

605K

Cost per run

$0.39

$1.61 est.

Cost per case

$0.0026

$0.011 est.

Time per case

3m 35s

1m 3s

Quantization

FP8

BF16

Harness

openrouter

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.

DeepSeek V4.1 FlashSpark X2.6 4B

Every test case

153 test cases

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

Agents80

Structured Extraction160

Coding110

Grounded Answers140

Instruction Following140

Business Logic150

Security & permissions112

Diagnosis & investigation10

90 won by DeepSeek V4.1 Flash2 won by Spark X2.6 4B61 tied