Breaking

OpenAI's GPT-5.6 Sol/Terra/Luna Added to Agent Arena

July 9, 2026 at 14:24 EDT

Foundation Models
AI Agents
Research & Papers

Arena.ai said on July 9, 2026 that it had added OpenAI's new GPT-5.6 models — Sol, Terra and Luna — to Agent Arena, a leaderboard that compares models on real-world agentic tasks. The move makes it possible to benchmark OpenAI's latest models as autonomous agents right away.

July 9, 2026 · Agent Arena

OpenAI's GPT-5.6 Family Enters the Agentic Arena

The new Sol, Terra and Luna models are added to a leaderboard that ranks AI on millions of real-world, long-horizon agentic tasks — with community-driven scores still being computed.

91.9%

Sol Ultra — state-of-the-art on Terminal-Bench 2.1

947K

Sessions logged across 32 models on the Arena

Durability tiers: Sol, Terra, Luna

Output price per 1M tokens

Column height is proportional to output cost across the three tiers

$30

Sol

Flagship

$15

Terra

Balanced

Luna

Fast, low-cost

Terra delivers GPT-5.5-class performance at roughly half the price.

How the Agent Arena ranks a model

Real tasks

Community-submitted, long-horizon

→

Tools

Web · filesystem · terminal

→

Live signals

Reliability · completion · steerability

→

Ranking

Net Improvement score

Current leader: Claude Fable 5 (High) at a Net Improvement of 14.10%. GPT-5.6 is a new entrant — its scores are still being computed.

Optimism

Terra's cost efficiency and Sol's agentic coding and long-horizon performance draw strong early interest, backed by "Thinking/High" resilience on bash recovery.

Caution

Concerns persist over sustainability — the "temporary" limited preview and government-gated access — with final leaderboard placement still unresolved.

Continue reading

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