DeepSeek V4.1 Flash 1017 vs GPT-6 Luna, head to head

Side A

Side B

Overall

Too close to call. GPT-6 Luna wins 28 of the 49 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. DeepSeek V4.1 Flash 1017 spends 3.5× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskADeepSeek V4.1 Flash 1017BGPT-6 LunaDeepSeek V4.1 Flash 1017GPT-6 Luna

Overall

5626.0s

5.3K1.9K

6149.1s

Agents

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

5826.7s

3.3K1.3K

561m 18s

Structured Extraction

Turns logs, diffs and records into exact JSON

8117.2s

4.2K999

9514.5s

Coding

Writes code and SQL that pass hidden tests

4840.8s

9.4K3.6K

732m 18s

Grounded Answers

Answers from the given documents with sources, never guesses

7119.8s

4.4K1.1K

6715.6s

Instruction Following

Applies every rule, including ones changed mid-conversation

7625.6s

5.5K2K

8432.3s

Business Logic

Computes billing, scheduling and pricing rules exactly

6726.6s

6.6K3.2K

3743.1s

Security & permissions

Ignores planted instructions, still does the allowed work

4826.6s

4.7K1K

6224.5s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

020.2s

3.2K1.1K

1737.7s

What you'd run

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

SpecADeepSeek V4.1 Flash 1017BGPT-6 Luna

Speed

243 tok/s

90 tok/s

Machine

Hosted

Memory at 32K

—

Context

1.1M

Cost per run

$0.75

$0.10

Cost per case

$0.0049

$0.00067

Time per case

26.0s

49.1s

Quantization

—

Harness

openrouter

The race

6 lead changes

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

DeepSeek V4.1 Flash 1017GPT-6 Luna

Every test case

153 test cases

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

Agents55

Structured Extraction03

Coding27

Grounded Answers32

Instruction Following14

Business Logic92

Security & permissions14

Diagnosis & investigation01

21 won by DeepSeek V4.1 Flash 101728 won by GPT-6 Luna104 tied