AI Model Rankings: Claude-Fable-5 Holds the Crown as Chinese Contenders Inch Up

claude-fable-5 is still the model to beat. The latest weekly rankings from Arena show Anthropic’s flagship holding the top spot, with a one-point ELO bump that barely registers as a change.

The top five models are all Claude variants, separated by fewer than 15 ELO points — well within the statistical margin of error. US models swept the top ten slots. The only notable movement just outside that group came from Google’s gemini-3.5-flash, which climbed two spots to 11th. glm-5.1 dropped three places to 22nd, the biggest downward shift in the overall chart.

Chinese models held their positions in the middle of the pack. Alibaba’s qwen3.7-max-preview is the highest-ranked Chinese entry at 15th, with an ELO of 1475, up one spot. Baidu’s ernie-5.1 sits at 27th (1468 ELO), Xiaomi’s mimo-v2.5-pro at 29th (1466 ELO). Neither gained nor lost ground. None cracked the top 10, but none slipped out of the top 30 either.

Anthropic’s dominance is most visible in the coding category, where its models hold the top nine slots. claude-fable-5 leads at 1563 ELO, just 10 points ahead of claude-opus-4-7-thinking — again within error-margin territory. The biggest loser here was glm-5.2 (max), which dropped eight spots to 21st after shedding 10 ELO points.

Chinese models have a meaningful presence in coding. Alibaba’s qwen3.7-max-preview is 10th at 1526 ELO. Xiaomi’s mimo-v2.5-pro sits at 17th (up one, 1520 ELO). Moonshot AI’s kimi-k2.6 is 28th (1512 ELO). Baidu’s ernie-5.1 and Alibaba’s qwen3.7-plus round out 29th and 30th. Five Chinese models in the coding top 30 is not dominance, but it’s a signal.

In math, claude-opus-4-6-thinking leads at 1518 ELO. GPT-5.5 made the biggest jump in the entire report, climbing four spots to 8th. glm-5.2 (max) bounced back eight spots to 24th in math after its coding slide. Moonshot’s kimi-k2.6 inched up two spots to 13th, and qwen3.7-max-preview rose one to 10th.

The long-text category flipped: claude-fable-5 moved from second to first at 1524 ELO, pushing claude-opus-4-6-thinking to second. The gap is exactly one ELO point. Xiaomi’s mimo-v2.5-pro crept up one spot to 14th, and gpt-5.5-high jumped three to 16th.

Google and OpenAI both have new models in the pipeline. The current rankings may look different once those ship.