GPT-6.1 Sol vs GPT-6.1 Terra

The two GPT-6.1 models, a month apart, and OpenAI gives each a different job: Sol for complex coding and professional work at near-Astra quality, Terra for fast, efficient work on images, screens and writing. On the table they share more than that positioning suggests, which makes the rows that differ easier to weigh.

Where they match

The same context window and maximum output, the same long-context rule, and the same cache-read discount — both bill a cached read at 0.05x their input rate, deeper than any other current model. Their request and token limits per minute match tier for tier on the table OpenAI's guide shares across Astra, Sol and Terra. Both offer Standard, Batch, Flex and Fast, both need the Responses API for tool calling, and both support only US and EU data residency, Fast included.

Where Terra is cheaper

Every price cell. Input, cached input, cache writes and output are all lower on Terra, in both context bands and at every tier both publish. For a workload Terra handles well, there is no row where staying on Sol saves money.

What only one of them has

Sol has Ultrafast; Terra has no Ultrafast row. Terra accepts none reasoning effort, which Sol doesn't — both run to max with medium as the default. Terra's knowledge cutoff is newer. And Terra's rate-limit table publishes a batch-queue column, which Sol's lacks, so only Terra has a published Batch queue limit.

Inside Codex

Sol is the CLI's default; Terra needs CLI 0.159.0 or later and is offered from Plus upward. OpenAI's five-hour estimates give Terra the larger message range on Plus, and its credit rate is lower — the same work draws less of a plan or a credit balance.

How to decide

Keep the hardest coding on Sol, which OpenAI positions closest to Astra. Try Terra on screen-reading, visual and writing-heavy tasks, and on anything that needs none effort, at the same effort level — and measure turns per finished task rather than the per-token price.