ULTRABENCH Created by Iternal Technologies, the AI Strategy Experts

Intelligence vs Time

Every ranked LLM plotted by intelligence index against its release date, with the best-so-far line, open vs closed weights and the yearly pace.

Best so far: 93.7, GPT-6 Astra (2026/09/03). Open weights: 84.3, Kimi K3 (2026/07/16); closed weights: 93.7; a gap of 9.4 points. Over the last 12 months (Dec 2025 – Nov 2026) the best so far rose at +18.8 index points a year (a least-squares line through its value at each month's end).

Best-so-far records
ModelReleasedWeightsIndexGain
GPT-6 Astra OpenAI2026/09/03Closed93.7+2.2
Claude Fable 5.1 Anthropic2026/09/01Closed91.5+0.4
Claude Fable 5 superseded Anthropic2026/06/09Closed91.1+5.6
Claude Opus 4.8 superseded Anthropic2026/05/27Closed85.5+0.5
Gemini 3.1 Pro Google2026/02Closed85.0+7.6
GPT-5.2 Pro superseded OpenAI2025/12/10Closed77.4+2.5
GPT-5.1-Codex-Max OpenAI2025/11/19Closed74.9+0.7
Qwen3-VL-235B-A22B-Thinking superseded Alibaba2025/05/14Open74.2+10.4
gemini-2.5-pro-preview-03-25 Google2025/05/07Closed63.8+3.8
Phi-4-reasoning Microsoft2025/04Open60.0+6.5
Gemini 2.5 Pro Google2025/03Closed53.5+2.5
OpenAI o1 superseded OpenAI2024/12/17Closed51.0+18.1
Qwen2.5 72B Instruct Alibaba2024/09/19Open32.9+4.5
Qwen2.5-32B-Instruct superseded Alibaba2024/09Open28.4+5.1
Llama-3.1-405B-Instruct Meta2024/07/23Open23.3+2.2
GPT-4 Turbo OpenAI2024/04/09Closed21.1+10.1
GPT-4 OpenAI2023/05/28Closed11.0—