Agnes 2.5 Pro Beta scores 49 on Intelligence Index
Singapore-based Agnes AI's new beta checkpoint gains nine points on the Artificial Analysis Intelligence Index, mostly on agentic tasks, putting it just behind Gemini 3.5 Flash and GPT-5.6 Luna while using about twice the output tokens.
Agnes 2.5 Pro Beta, a beta checkpoint of Agnes AI's next 2.5 Pro release, scores 49 on the Artificial Analysis Intelligence Index, up from 40 for Agnes 2.5 Pro Alpha, according to an evaluation by Artificial Analysis. That puts it just behind Gemini 3.5 Flash (high) and GPT-5.6 Luna (max), both at 52, and ahead of MiniMax-M3 at 45.
Most of the gain is on agentic work. The model's Agentic Index score rose from 25 to 44. τ³-Banking nearly tripled, from 12% to 36%. On GDPval-AA v2 its Elo went from 1171 to 1456, against a human baseline of 1000. That is ahead of MiniMax-M3 (1384) and behind GPT-5.5 (xhigh, 1489) and Gemini 3.7 Flash (high, 1527).
Reasoning scores improved less. Humanity's Last Exam went from 34% to 38%, GPQA Diamond from 88% to 91% and CritPt from 11% to 16%.
The AA-Omniscience score improved from -25 to -11, but Artificial Analysis says this came from the model declining to answer, not from better accuracy. The Beta attempted 45% of questions, compared with 94% for the Alpha. Its hallucination rate fell from 88% to 33%, but its accuracy halved, from 33% to 17%.
The gains also cost more compute. The model used 50k output tokens per Intelligence Index task, up from 24k for the Alpha and more than Qwen3.8 27B (xhigh, 47k) uses.
Agnes 2.5 Pro Beta is a reasoning model that takes text and image input and produces text. It has a 1M-token context window and up to 65k output tokens. It costs $0.10 per million input tokens, $0.30 per million output tokens and $0.01 per million cached tokens, and is available only through Agnes AI's own API.