GPT-6.1 Terra's price cut, recomputed for a cache-heavy agent session

GPT-6.1 Terra is OpenAI’s new mid-tier model, priced at $1 per million fresh input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. Compared with GPT-5.6 Terra, those rates are lower by 50% for fresh input, 75% for cached input, and 58% for output. Because coding agents repeatedly send cached context, the cache discount can make their savings especially large. In an illustrative session with 10 million input tokens, 97% cached, and 200,000 output tokens, Terra costs $1.79, versus $4.94 for GPT-5.6 Terra and $3.57 for either GPT-6.1 Sol or Claude Sonnet 5.5. GPT-6 Luna is cheaper still, at $0.23. These estimates assume equal token use across models and exclude cache-writing fees.

A key caveat is a sharp pricing threshold: requests exceeding 272,000 input tokens incur twice the normal input and cache rates and 1.5 times the output rate for the entire request, despite Terra supporting up to 922,000 input tokens. A 300,000-token prompt can therefore cost more than twice as much as a 250,000-token prompt. Independent tests put Terra’s cost per task at $0.40, down from $1.40 for its predecessor; its intelligence score is four points below Sol and ten above Luna, while its coding score trails Sol. The practical case for switching depends on real task quality, cache-hit rates, and prompt size: teams should test their workloads, monitor token use, and split or trim requests nearing the threshold.