Anthropic and OpenAI released their new model versions on the same day, and everyone is being very honest with their actions. On one hand, they call on peer companies to slow down and push for legislation to set thresholds; on the other, they race to release new models.
The Sol and Luna released by OpenAI this time are additions to the GPT-6 family, and compared with 5.6, the mid-tier Terra is missing.
On pricing, for comparison, listed directly as follows (all per million Tokens input/output):
- GPT-6 Astra input $10 output $50
- GPT-6 Sol input $2 output $10
- GPT-6 Luna input $0.1 output $0.5
The prices of GPT-6 Sol and Luna are both 50% cheaper than GPT-5.6, and DS Flash is no longer the king of cost-effectiveness.
From the official line, in terms of overall model performance, GPT-6 Sol is benchmarked against Claude Opus 5, but who knew Anthropic would release Opus 5.5 today; overall, Sol is probably a beat behind, though its price is cheap enough to be hard to resist.
The highlights of GPT-6 Sol this time are as follows:
- Strengthened the factual accuracy of answers; in internal testing, the number of errors was half that of 5.6, approaching Astra's level
- In coding performance, it focuses on larger-scale code tasks, which aligns with Opus 5.5's goal
- In Computer use capability, it keeps pace with Astra, achieving solid control at lower cost
- Optimized the response approach, with less filler and higher information density in answers
Currently GPT‑6 Sol and GPT‑6 Luna are fully available. Who's next? Gemini Pro?