ItInsightsWorld Breakdown of GPT-6.1 Terra: Capability Profile, Rate Card, and Endpoint Map
Before a new frontier model gets budget in your stack, ItInsightsWorld has pulled the published model card for GPT-6.1 Terra (gpt-6.1-terra) apart into the pieces an evaluation actually needs: what it claims to do well, how its reasoning controls work, where the ceilings sit, and how the meters run. The description on offer centres on visual inference, accurate computer interaction and lucid natural-language writing, pitched as fast and efficient, with a side-by-side comparison and a Playground trial provided for anyone who wants to check those claims first-hand.
Snapshot of the headline attributes
- Reasoning tier: Highest
- Speed rating: Fast
- Listed rates: $1 input and $5 output per million tokens
- Accepted input: Text, Image
- Returned output: Text
Workloads the product page points at
Rather than a generic capability list, the entry frames this model around concrete jobs: aligning product photos with catalog entries, reading technical drawings, operating application screens, and creating or revising text.
Reasoning settings, ceilings and reference material
The reasoning.effort field accepts none, low, medium (the default), high, xhigh and max. Tool calling is only available through the Responses API, while prompting plus migration guidance is written up in the GPT-6 guide. The published envelope is a 1,050,000 context window with 128,000 max output tokens, the knowledge cutoff is listed as Jun 01, 2026, and reasoning token support is confirmed on the card as well.
How the billing meters run
Chargeable amounts are derived from token counts or other metrics tied to the model type. Tool-specific models such as search and computer use carry an additional fee per tool call, itemised on the pricing page.
Text token rates per 1M tokens
| Line item | Per 1M tokens |
|---|---|
| Input | $1.00 |
| Cached input | $0.05 |
| Cache writes | $1.25 |
| Output | $5.00 |
Put differently, cached input is billed at 5% of the uncached input token rate.
The cache-write numbers shown correspond to 30-minute retention. Under Standard cache writes, the rate is $1.25 per million tokens for 30-minute retention and rises to $2.50 for 12-hour retention, quoted before long-context adjustments.
Long-context billing takes hold once a prompt passes 272K input tokens: the entire request is then invoiced at 2x input and cache rates and 1.5x output.
Workloads that can tolerate latency benefit from Batch and Flex pricing, which runs 50% lower than Standard.
Where regional processing is available, it is applied as a 10% premium.
Fast mode is priced at 2x Standard.
Which modalities are covered
| Modality | Support |
|---|---|
| Text | Input and output |
| Image | Input only |
| Audio | Not supported |
| Video | Not supported |
Endpoint paths to wire against
| Endpoint | Path |
|---|---|
| Live | v1/live/sessions |
| Chat Completions | v1/chat/completions |
| Responses | v1/responses |
| Realtime | v1/realtime |
Read as a whole, the card positions GPT-6.1 Terra as a high-reasoning model that takes text and images but returns only text, so the economics of any deployment hinge on cache reuse and prompt length far more than on the headline rate. Prepared by James Hartley for ItInsightsWorld (IT Insights).