Breaking

OpenAI's GPT-6 Sol Reaches 7.68% Net Improvement in Agent Arena at $0.74 per Task

September 25, 2026 at 15:52 EDT

Foundation Models
AI Agents
Software Dev & Coding

OpenAI's GPT-6 Sol (Max) has entered the upper tier of agentic benchmarks on cost as much as raw score, posting a 7.68% net improvement on Agent Arena at a $0.74 median cost per task. The result puts it 0.60 percentage points behind Claude Fable 5 (High), while costing about 57% less per task.

Sep 25, 2026 | OpenAI

GPT-6 Sol turns cost into the story

GPT-6 Sol (Max) posted a 7.68% net improvement at a $0.74 median task cost, landing close to Claude Fable 5 (High) while costing much less per task.

Net improvement

7.68%

Median task cost

$0.74

Gap to Fable 5 High

0.60 pp

Sol sessions

3,985

Median Cost Per Task

Each block is about $0.25, so column height reflects cost.

$0.74

GPT-6 Sol (Max)

$1.71

Claude Fable 5 (High)

Why it matters

Lower per-task cost gives teams more room to iterate, and the model leads the visible steerability signal at 15.70%.

What to watch

Only 3,985 sessions are listed so far, and users report mixed results on complex debugging and rollout consistency.

Not a raw-score crown

Claude Fable 5.1 (Max) still leads overall at 13.80%, while GPT-6 Sol's case rests on cost, steerability and more data to come.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.