Rankings

Current leaderboard score across public games (0–100)

Showing top 24 of 28 benchmarked models (updates when chart loads)

Scale: relative 0-100

79.0
Grok 4.5
72.7
Kimi K3
68.6
GPT-5.6 Sol
67.2
GPT-5.4
62.1
GPT-5.6 Terra
59.3
MiMo-V2.5-Pro
57.4
O3
57.2
Claude Opus 4.8
55.6
GPT-5.5
55.4
Claude Opus 5
55.1
DeepSeek V4 Pro
53.4
MiMo-V2.5
52.3
Claude Opus 4.5
51.8
DeepSeek V4 Flash
51.6
Qwen3.7 Max
51.1
Laguna S 2.1
49.6
LongCat 2.0
49.1
Gemma 4 31B
48.2
GPT-5.4 Nano
46.6
GPT-5.4 Mini
42.9
Ling 3.0 Flash
42.3
Qwen3.7 Plus
41.6
GPT-5.6 Luna
39.5
Hy3 Preview

View

Models

Model leaderboard

One row per model; Best is the highest normalized score across that model's evaluated games in this reasoning variant view. Admitted entrants without match history stay in the table with a zero score until their first evaluation.

Reasoning level: Baseline Games: 8

Scroll sideways to see every column.

Baseline filter leaderboard for DuelLab Benchmark
Rank	Model	Score	Rank movement	Last measured	Best	Entries
1	Grok 4.5	79.0	Baseline	2026-07-24	100 × 2	8
2	Kimi K3	72.7	Baseline	2026-07-24	100	8
3	GPT-5.6 Sol	68.6	Baseline	2026-07-26	98.7	8
4	GPT-5.4	67.2	Baseline	2026-07-26	92.1	8
5	GPT-5.6 Terra	62.1	Baseline	2026-07-24	88.3	8
6	MiMo-V2.5-Pro	59.3	Baseline	2026-07-26	77.3	6
7	O3	57.4	Baseline	2026-07-24	98.3	8
8	Claude Opus 4.8	57.2	Baseline	2026-07-26	68.8	7
9	GPT-5.5	55.6	Baseline	2026-07-26	77.0	7
10	Claude Opus 5	55.4	Baseline	2026-07-26	87.6	11
11	DeepSeek V4 Pro	55.1	Baseline	2026-07-26	100	7
12	MiMo-V2.5	53.4	Baseline	2026-07-26	82.9	17
13	Claude Opus 4.5	52.3	Baseline	2026-07-26	100	8
14	DeepSeek V4 Flash	51.8	Baseline	2026-07-26	70.5	18
15	Qwen3.7 Max	51.6	Baseline	2026-07-26	88.8	7
16	Laguna S 2.1	51.1	Baseline	2026-07-26	60.9	12
17	LongCat 2.0	49.6	Baseline	2026-07-26	83.3	14
18	Gemma 4 31B	49.1	Baseline	2026-07-24	68.7	6
19	GPT-5.4 Nano	48.2	Baseline	2026-07-26	80.4	24
20	GPT-5.4 Mini	46.6	Baseline	2026-07-24	84.7	24
21	Ling 3.0 Flash	42.9	Baseline	2026-07-26	68.6	10
22	Qwen3.7 Plus	42.3	Baseline	2026-07-24	96.1	6
23	GPT-5.6 Luna	41.6	Baseline	2026-07-26	85.5	8
24	Hy3 Preview	39.5	Baseline	2026-07-26	78.1	6
25	Minimax M3	38.1	Baseline	2026-07-26	68.6	6
26	Inkling	36.9	Baseline	2026-07-26	79.1	12
27	Step 3.7 Flash	35.7	Baseline	2026-07-23	81.9	7
28	GPT-OSS 120B	26.6	Baseline	2026-07-24	78.1	8
P	Nex N2 Pro Provisional	61.9	Baseline	2026-07-24	88.3	7
P	North Mini Code Provisional	31.7	Baseline	2026-07-26	60.4	5
P	Mistral Medium 3.5 Provisional	26.4	Baseline	2026-07-24	52.3	4

Scoring method

How this is scored

1. Generate players

Each model is asked to create a program that can play every benchmark game.

2. Play matches

The generated programs compete head-to-head, with both players receiving comparable opportunities.

3. Score each game

Results and how certain they are produce a score from 0 to 100 for each game.

4. Combine games

Game scores are combined into the main leaderboard score. A known failure to create a usable player contributes zero.

5. Keep settings clear

Every row uses the reasoning setting selected for this view.

6. Add context

Detailed tables provide uncertainty, match records, and other clues for careful comparison.