DeepSeek V4.1 Flash 1017 vs GLM 5.3 Flash, head to head

Side A

Side B

Overall

DeepSeek V4.1 Flash 1017 is clearly better. It wins 58 of the 75 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 1.9× as many thinking tokens per test case.

Thinking tokensAnswer tokens

TaskADeepSeek V4.1 Flash 1017BGLM 5.3 FlashDeepSeek V4.1 Flash 1017GLM 5.3 Flash

Overall

5626.0s

5.3K2.8K

3555.5s

Agents

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

5826.7s

3.3K607

1037.0s

Structured Extraction

Turns logs, diffs and records into exact JSON

8117.2s

4.2K2.7K

4847.2s

Coding

Writes code and SQL that pass hidden tests

4840.8s

9.4K5.3K

541m 52s

Grounded Answers

Answers from the given documents with sources, never guesses

7119.8s

4.4K2.9K

3756.8s

Instruction Following

Applies every rule, including ones changed mid-conversation

7625.6s

5.5K2K

4340.8s

Business Logic

Computes billing, scheduling and pricing rules exactly

6726.6s

6.6K5.8K

331m 18s

Security & permissions

Ignores planted instructions, still does the allowed work

4826.6s

4.7K963

5525.2s

Diagnosis & investigation

Chooses diagnostic checks, repairs supported causes and verifies recovery

020.2s

3.2K563

027.0s

What you'd run

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

SpecADeepSeek V4.1 Flash 1017BGLM 5.3 Flash

Speed

243 tok/s

60 tok/s

Machine

Hosted

Memory at 32K

—

Context

Cost per run

$0.75

$0.19

Cost per case

$0.0049

$0.0013

Time per case

26.0s

55.5s

Quantization

—

FP8

Harness

openrouter

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 Flash 1017GLM 5.3 Flash

Every test case

153 test cases

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

Agents110

Structured Extraction92

Coding65

Grounded Answers102

Instruction Following81

Business Logic92

Security & permissions55

Diagnosis & investigation00

58 won by DeepSeek V4.1 Flash 101717 won by GLM 5.3 Flash78 tied