Agnes 2.5 Pro Beta Enters Frontier Tier with Agentic Gains at Doubled Token Cost

Singapore's Agnes AI released the omni-modal foundation model Agnes 2.5 Pro Beta. Per Artificial Analysis, it scores 49 on the Intelligence Index, up 9 points from Alpha, placing it in the frontier tier just behind Gemini 3.5 Flash and leading GPT models, with the evaluator noting the score approaches GPT-5.6.

Confirmed

• Intelligence Index score of 49, up 9 points from Alpha's 40, entering the frontier tier
• The gain is driven mainly by agentic capability: Agentic Index rose from 25 to 44
• Elo rating of 1456 on the GDPval-AA v2 benchmark, approaching GPT-5.5
• Tasks averaged about 50k output tokens, more than double Alpha's 24k and higher than comparable Qwen3 models
• Artificial Analysis noted part of the score increase relates to changes in the model's refusal behavior

Why it matters

Agnes 2.5 Pro Beta illustrates a "compute-for-intelligence" path: big agentic gains come with doubled token consumption, meaning significantly higher real inference costs. For token-billed deployments, weighing performance against cost is the key consideration.

2026-08-28 ~ 2026-08-28 · 5 related posts

• [source] Agnes 2.5 Pro Beta Joins Frontier Tier with Major Agentic Gains, Higher Cost — ArtificialAnlys · 2026-08-28
• [source] Agnes 2.5 Pro Beta Agent Score Soars to Elo 1456, Near GPT-5.5 — ArtificialAnlys · 2026-08-28
• [source] Agnes 2.5 Pro Beta Uses 50k Output Tokens; Score Rise Driven by Abstention — ArtificialAnlys · 2026-08-28
• Singapore's Agnes 2.5 Pro Beta scores 49, nears GPT-5.6 Luna — yacineMTB · 2026-08-28

1 near-duplicate retellings: ArtificialAnlys