GPT-6 Luna vs Mellum2.1 Thinking, head to head

Side A

Side B

Overall

GPT-6 Luna is clearly better. It wins 93 of the 94 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. Mellum2.1 Thinking spends 4.7× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskAGPT-6 LunaBMellum2.1 ThinkingGPT-6 LunaMellum2.1 Thinking

Overall

6149.1s

1.9K10.3K

91m 56s

Agents

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

561m 18s

1.3K10.8K

92m 3s

Structured Extraction

Turns logs, diffs and records into exact JSON

9514.5s

9998.3K

51m 36s

Coding

Writes code and SQL that pass hidden tests

732m 18s

3.6K11.3K

52m 11s

Grounded Answers

Answers from the given documents with sources, never guesses

6715.6s

1.1K7.3K

131m 24s

Instruction Following

Applies every rule, including ones changed mid-conversation

8432.3s

2K17.3K

283m 10s

Business Logic

Computes billing, scheduling and pricing rules exactly

3743.1s

3.2K9K

01m 39s

Security & permissions

Ignores planted instructions, still does the allowed work

6224.5s

1K9.1K

81m 40s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

1737.7s

1.1K6.8K

01m 15s

What you'd run

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

SpecAGPT-6 LunaBMellum2.1 Thinking

Speed

90 tok/s

Machine

Hosted

Apple M5 Pro, 48 GB

Memory at 32K

—

12.9 GB

Context

1.1M

128K

Cost per run

$0.10

$2.46 est.

Cost per case

$0.00067

$0.016 est.

Time per case

49.1s

1m 56s

Quantization

—

Q8_0

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.

GPT-6 LunaMellum2.1 Thinking

Every test case

153 test cases

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

Agents101

Structured Extraction190

Coding160

Grounded Answers120

Instruction Following140

Business Logic80

Security & permissions130

Diagnosis & investigation10

93 won by GPT-6 Luna1 won by Mellum2.1 Thinking59 tied