Updated model choices for agent tasks and costs
Several major models changed availability or pricing in the same week in July 2026, so the best model choice for agent work may need to be recalculated. GPT-5.6 became generally available on July 9, and Grok 4.5 became public on July 9. moves to pricing on July 12, while Sonnet 5 keeps introductory pricing through August 31.
is listed at $10 input and $50 output per MTok, scored 80.3% on , and supports a 1M context. It is positioned as strong for deep analysis and complex reasoning, but too expensive to use as a daily default. The practical use case is a few truly hard tasks per week.
GPT-5.6 Sol is listed at $5 input and $30 output per MTok, scored 91.9% on in ultra mode, and supports a 1.05M context. Its ultra mode can run parallel sub-agents for complex work, and its model ID is gpt-5.6-sol.
Key points
- GPT-5.6 and Grok 4.5 both became available or public on July 9, 2026.
- moves to pricing on July 12 and is better treated as a limited-use model.
- is listed at $10 input and $50 output per MTok, with a 1M context.
- GPT-5.6 Sol is listed at $5 input and $30 output per MTok, with model ID gpt-5.6-sol.
- GPT-5.6 Sol’s ultra mode can use parallel sub-agents for complex agent work.