Breaking

OpenAI's GPT-6 Sol and Luna Put Cost Per Task at the Center of the AI Model Race

September 23, 2026 at 14:46 EDT

Foundation Models
AI Agents
Software Dev & Coding

OpenAI's new GPT-6 Sol and GPT-6 Luna are turning the latest foundation-model race into a cost-per-task contest, not just a leaderboard contest. A same-prompt 3D lava-lamp run shows why: Claude Opus 5.5, GPT-6 Astra, GPT-6 Sol and GPT-6 Luna all qualified, but their bills ranged from $1.27 to under $0.01.

September 23, 2026 · Foundation models

Same Prompt, Very Different Bills

A one-shot 3D lava lamp task put four frontier models on the same brief. All qualified, but cost and time split sharply by tier.

$1.27

Claude Opus 5.5 lava lamp run

$0.08

GPT-6 Sol lava lamp run

<$0.01

GPT-6 Luna lava lamp run

50%

API price cut versus GPT-5.6 promo pricing

Lava Lamp Cost, Drawn to Scale

Each stacked segment is about $0.10 of generation cost; Luna is shown as a sub-cent sliver.

$1.27

Claude Opus 5.5

$0.63

GPT-6 Astra

$0.08

GPT-6 Sol

<$0.01

GPT-6 Luna

Why builders like it

Lower token prices give teams more room to iterate, run routine agent steps, and push high-volume work toward cheaper tiers.

What still needs checking

A pass/fail UI result does not judge visual taste, and model routing now depends on task difficulty, effort settings, caching, and review needs.

The routing question

Hardest work
Astra or Opus 5.5

→

Coding agents
Sol for cost balance

→

Volume tasks
Luna for scale

Continue reading

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