September 2026 AI Large Language Model Rankings: Claude Fable 5.1 Tops the List

Overview

In September 2026, the competitive landscape in the field of large AI models underwent significant changes. Claude Fable 5.1, released by Anthropic, topped multiple authoritative evaluations, establishing its industry-leading position. This result broke the previous situation where OpenAI's GPT-6 series dominated the top rankings, demonstrating a new round of breakthroughs in complex reasoning and general capabilities for closed-source models.

Currently, there is no unified standard ranking system in the industry. Therefore, this list comprehensively combines two highly representative evaluation dimensions: the Artificial Analysis Intelligence Index and LMArena (formerly Chatbot Arena), presenting the latest technological trends from both objective benchmark testing and user subjective preference perspectives.

Comprehensive Ranking Dual-Dimension Analysis

The AA Intelligence Index aggregates scores through 10 standardized evaluations covering programming, mathematics, science, and reasoning, focusing on measuring the model's objective core capabilities. Data updates as of September 13, 2026, show that Claude Fable 5.1 ranks first with a high score of 53 points, tying for the top tier with GPT-6 Astra across different modes. Close behind are Claude Opus 5 (max) and Claude Fable 5. Anthropic's family secures four of the top five spots, showcasing strong technological iteration capabilities.

On the other hand, LMArena's Elo rating based on global user anonymous blind tests reflects perceived quality in real-world usage scenarios. Update data from September 11 shows that Claude Fable 5 ranks first with an Elo score of 1506 points, while Claude Opus 4.6 (high) takes second place with a narrow margin. Notably, Muse Spark 1.2 (xhigh) from Facebook AI Research Lab enters the top five, highlighting the rise of emerging forces within the open-source ecosystem.

Dynamic Rankings for Individual Capability Tracks

In specialized capability tracks, competition among manufacturers is fierce, with new models frequently appearing on the lists. In terms of coding ability, DeepSeek V4.1 Flash, a newly listed model, performs impressively, ranking first with 3471 points, demonstrating the rapid catch-up of the Chinese developer community in the field of programming large models. Google Deep Mind's Gemini 3 Deep Think and DeepSeek V3.2 Speciale rank second and third, respectively.

The mathematical ability leaderboard is led by xAI's Grok 4 Heavy with a perfect score of 100, followed by Google's Gemini 2.5 Deep Think and OpenAI's GPT-5 Codex in second and third place. Moonshot AI's Kimi K2 Thinking enters the top eight thanks to its Long-CoT technology, highlighting the effectiveness of long chain-of-thought strategies in complex mathematical reasoning.

Regarding Agent capabilities, Claude Fable 5 ranks first with a high score of 85 points, closely followed by OpenAI's o3-pro and Moonshot AI's Kimi K3. Alibaba's Qwen3.8-27B, a newly listed model, ranks fourth, while Google's Gemini 3.6 Flash also enters the top ten for the first time, reflecting that multi-agent collaboration capabilities have become a new focus of competition.

Parameter Comparison of Mainstream Models

From large-scale performance evaluations, Claude Fable 5.1 achieved the highest score of 65.00 points on the HLE benchmark, with Anthropic's Claude Mythos Preview following closely at 64.70 points. DeepSeek-V4.1-Flash scored 63.90 points to become the best open-source model and lead domestic models simultaneously; its free commercial license further promotes the development of the open-source ecosystem.

Zhipu AI's GLM-5.3 entered the top five with a score of 62.50 points, while Tencent AI Lab's Hy4 preview ranked with 55.40 points. The investment by leading domestic manufacturers in self-developed foundational models is gradually translating into measured results. Overall, the HLE scores of top-tier models are generally concentrated in the range above 50 points. The technological gap is narrowing, and comprehensive reasoning ability has become the key factor determining rankings.