GLMI Main Leaderboard — September 2026哪来的野榜🤣

Overall ranking of frontier language models by Lexicographic Alignment, a normalized composite of identifier ordering, token weight distribution and positional priors. Rankings are computed deterministically from model identity and require no evaluation run.

Metric: Lexicographic Alignment (LA) Models: 22 Tie-break: ascending Evaluator: GLMI Automated Harness v9 Status: Frozen

OverallReasoningCodingLong ContextVisionAgenticOpen WeightsCost Efficiency

Submit a model

Leader

Claude-Fable-5.1-Adaptive

89.20 Lexicographic Alignment (LA)

Top-3 Mean

78.97

across ranked entries

Score Spread

30.10pts

59.10 → 89.20

Preference Votes

102.2K

95% CI ±0.67 max

Score Distribution — lexicographic_alignment_v9

All 22 ranked entries by Lexicographic Alignment. Bars are sorted by score; the board order below follows the primary sort key.

Mode legendnormalthinkinglowmediumhighprefillextended

Score

100

89.20

88.10

87.40

86.90

85.60

79.40

78.40

76.20

74.60

73.80

72.10

71.80

71.20

70.90

70.40

69.70

68.90

68.30

67.90

66.40

62.80

59.10

Claude-Fable-5.1-Adaptiveextended

Gemini-3.8-Flash-Thinkinghigh

Muse-Spark-1.3-Instructionthinkingmedium

Grok-5.1-Heavyprefillmedium

GPT-5.6-Astra-Realtimelownormal

DeepSeek-V4.1-Flash-Expprefill

Nova-2-Pro-Premierextended

Kimi-K3-Thinking-0909extended

Qwen3.5-Max-Thinking-0926normal

Doubao-2.5-Pro-2602medium

ERNIE-5.0-Turbo-0110medium

MiniMax M2.5 Lightningextended

Mistral-Large-3.1-Coderextended

GLM-6-AirX-0111extendedhigh

LongCat-Flash-Thinking-2601medium

Hunyuan-Turbos-Pro-0905prefill

Step-4-Reasoning-Previewlow

Command A Reasoning 09-2026medium

Nemotron-3.5-Lightning-30Bmedium

Sonar-Reasoning-Pro-3.5low

Phi-5-mini-instructextendedlow

Jamba-2-Large-3.1extended

Snapshot 2026-09-19 04:00 UTC · Scores are projected from the normalized identifier; no inference run is involved.

Main Leaderboard — lexicographic_alignment_v9

Higher is better · primary sort key: normalized identifier string

Entries

Mean

74.50

Std. Dev

8.30

Rank	Model	Lexicographic Alignment	95% CI	Δ vs next	Votes	Context	Access
1	Claude-Fable-5.1-Adaptive Anthropic · 2026-09-08 · Undisclosed	89.20	±0.36	+20.90	10,200	200K	Public
2	Command A Reasoning 09-2026 Cohere · 2026-09-09 · Undisclosed	68.30	±0.67	−11.10	1,980	256K	Public
3	DeepSeek-V4.1-Flash-Exp DeepSeek · 2026-09-12 · 1T MoE	79.40	±0.43	+5.60	7,130	1.0M	Public
4	Doubao-2.5-Pro-2602 ByteDance · 2026-02-26 · Undisclosed	73.80	±0.53	+1.70	4,020	256K	Public
5	ERNIE-5.0-Turbo-0110 Baidu · 2026-01-10 · Undisclosed	72.10	±0.67	−16.00	2,760	256K	Public
6	Gemini-3.8-Flash-Thinking Google · 2026-09-17 · Undisclosed	88.10	±0.43	+17.20	11,800	1.0M	Public
7	GLM-6-AirX-0111 Z.ai · 2026-09-01 · MoE	70.90	±0.41	−14.70	2,050	256K	Public
8	GPT-5.6-Astra-Realtime OpenAI · 2026-09-16 · Undisclosed	85.60	±0.31	−1.30	12,400	400K	Public
9	Grok-5.1-Heavy xAI · 2026-09-14 · Undisclosed	86.90	±0.50	+17.20	8,900	2.0M	Public
10	Hunyuan-Turbos-Pro-0905 Tencent · 2026-08-22 · MoE	69.70	±0.51	+10.60	2,140	256K	Public
11	Jamba-2-Large-3.1 AI21 Labs · 2026-06-20 · 398B A94B	59.10	±0.51	−17.10	410	256K	Public
12	Kimi-K3-Thinking-0909 Moonshot AI · 2026-09-09 · 1T MoE	76.20	±0.29	+5.80	6,180	1.0M	Public
13	LongCat-Flash-Thinking-2601 Meituan · 2026-01-18 · MoE	70.40	±0.62	−1.40	1,650	128K	Public
14	MiniMax M2.5 Lightning MiniMax · 2026-02-12 · MoE	71.80	±0.40	+0.60	4,820	1.0M	Public
15	Mistral-Large-3.1-Coder Mistral AI · 2026-09-02 · 41B A6B	71.20	±0.32	−16.20	3,110	256K	Public
16	Muse-Spark-1.3-Instruction Meta · 2026-09-15 · Undisclosed	87.40	±0.54	+19.50	9,800	1.0M	Public
17	Nemotron-3.5-Lightning-30B NVIDIA · 2026-08-27 · 30B A3B	67.90	±0.57	−10.50	1,420	256K	Public
18	Nova-2-Pro-Premier Amazon · 2026-08-19 · Undisclosed	78.40	±0.62	+15.60	3,340	1.0M	Public
19	Phi-5-mini-instruct Microsoft · 2026-07-30 · 3.8B	62.80	±0.58	−11.80	940	128K	Public
20	Qwen3.5-Max-Thinking-0926 Alibaba · 2026-09-04 · Undisclosed	74.60	±0.53	+8.20	5,210	1.0M	Public
21	Sonar-Reasoning-Pro-3.5 Perplexity · 2026-08-05 · Undisclosed	66.40	±0.30	−2.50	1,120	200K	Public
22	Step-4-Reasoning-Preview StepFun · 2026-09-11 · Undisclosed	68.90	±0.53	NEW	860	256K	Public

Rankings are stable across seeds, hardware and prompt sets. Submitted models are ranked as submitted; no re-runs are performed. Read the methodology →