Open-Weight AI Models Trail Leaders by 12 Points

Private benchmark tests widen the measured gap, which may shape how investors view upcoming AI listings.

Updated on Nov. 3, 2026 in Artificial Intelligence

Open-Weight AI Models Trail Leaders by 12 Points

Live Poll

Do you trust large American AI companies to keep their promises to users?

Artificial Analysis, an independent firm that measures AI performance, overhauled its testing system and now keeps many of its tests secret. Under the new system, the best open-weight AI models trail the overall leader by 12 points, up from 3 points under the previous system.

Why it matters

Keeping tests secret makes benchmark gaming harder, since AI companies cannot train specifically for tests they can see. The revised rankings may also strengthen the investment case for Anthropic's planned IPO and a possible future OpenAI listing.

Artificial Analysis reports that Claude Haiku 5.5 outperformed every AI model with publicly released weights under the new system. The firm's overhaul aims to better reflect how people actually use the technology, though the report does not detail the specific benchmark methodology.

The players

Artificial Analysis

An independent firm that measures AI model performance through benchmark testing, recently overhauling its system to keep many tests secret.

Anthropic

An AI company whose Claude Opus 5.5 model ranks as the overall leader, and which is planning an IPO.

Sridhar Ramaswamy

A technology executive who said the open-model gap grew from 3 points to 12 when benchmarks went private.

OpenRouter

A platform that routes traffic to AI models, through which DeepSeek usage was tracked, representing only a slice of the AI model market.

The details

Benchmarks are standardized tests that score AI models on tasks such as reasoning or coding, letting developers compare systems. Artificial Analysis overhauled its system to better reflect real-world use, and many of its tests are now kept private rather than published. Because companies typically train models on known test material, hidden tests make it harder for a model to score well by memorizing the exam, which is why the gap between open-weight and closed models appears larger under the new system.

Under the previous testing system, the best open-weight models trailed the overall leader by 3 points.
On November 3, 2026, Business Insider published an article on the new rankings.

The Tech Race

The gap between the best open-weight models and the overall leader widened from 3 points to 12 points under the revised Artificial Analysis system. Chinese open-weight models, which trail the leader on this measure, face a competitive contest against closed American systems such as Claude Opus 5.5.

Developers and businesses comparing AI models will see different rankings under the new system, which may affect which tools they choose. The extraction does not state pricing or availability changes for any model.

The takeaway

Watch whether other benchmark firms adopt private testing and whether the open-weight gap persists on future rankings. The next verifiable marker is how Anthropic's planned IPO filing addresses model performance claims.