AIdataNewsGlobal LLMs: A Verifiable Read on GPT-6.1 Terra Capabilities and Pricing
This column reads the GPT-6.1 Terra (gpt-6.1-terra) brief as a decision sheet. The listed advantages are visual inference, accurate computer interaction, and lucid natural-language writing; the overview also describes the model as fast and efficient. Verification is positioned through comparison and a Playground trial.
Core specification snapshot
| Attribute | Recorded value |
|---|---|
| Reasoning | Highest |
| Speed | Fast |
| Price | $1 • $5 |
| Input | Text, Image |
| Output | Text |
The page frames practical work around four recurring tasks: aligning product photos with catalog entries, reading technical drawings, operating application screens, and creating or revising text. That framing is narrower than a general chat promise; it gives an evaluator something concrete to test in a workflow.
Control knobs and API dependencies
For reasoning.effort, the supported values are none, low, medium (default), high, xhigh, and max. The document also states that the Responses API is required for tool calling, so teams should not assume that tool use can be routed through every interface without checking that dependency. Prompting and migration guidance is available in the GPT-6 guide, making the documentation reference part of the adoption path rather than a peripheral footnote.
Context window, output ceiling, and knowledge date
The listed limits are 1,050,000 context window and 128,000 max output tokens, with a Jun 01, 2026 knowledge cutoff. Reasoning token support is also noted. For a column focused on measurable model behavior, the useful conclusion is simple: the model specification is large enough to support long document and multi-step tasks, but its knowledge boundary must be handled explicitly in any application that presents time-sensitive content.
Pricing architecture: tokens, caching, and surcharges
Costs depend on token counts or other metrics tied to the model type. Tool-specific models such as search and computer use also incur a fee per tool call; the referenced pricing page has details. This means an operating estimate should include both model generation volume and tool activity, not only prompt and completion tokens.
Text token rates
| Text tokens | Per 1M tokens |
|---|---|
| Input | $1.00 |
| Cached input | $0.05 |
| Cache writes | $1.25 |
| Output | $5.00 |
The cached input rate is 5% of the uncached input token rate. The cache-write figures shown correspond to 30-minute retention. Under Standard cache writes, the price is $1.25 per million tokens for 30-minute retention and $2.50 for 12-hour retention before long-context adjustments.
Retention, long-context, discount, and premium rules
| Condition | Pricing effect |
|---|---|
| Cache write, 30-minute retention | $1.25 per million tokens |
| Cache write, 12-hour retention | $2.50 before long-context adjustments |
| Prompt exceeds 272K input tokens | Full request billed at 2x input and cache rates and 1.5x output |
| Batch and Flex | 50% lower than Standard |
| Regional processing, where available | 10% premium |
| Fast mode | 2x Standard |
The rule for very large prompts is economically important: when a prompt exceeds 272K input tokens, the full request is billed at 2x input and cache rates and 1.5x output. In contrast, Batch and Flex pricing is 50% lower than Standard. Where regional processing is available, it adds a 10% premium, while Fast mode costs 2x Standard.
Modalities
| Modality | Support |
|---|---|
| Text | Input and output |
| Image | Input only |
| Audio | Not supported |
| Video | Not supported |
Endpoints
| Endpoint | Path |
|---|---|
| Live | v1/live/sessions |
| Chat Completions | v1/chat/completions |
| Responses | v1/responses |
| Realtime | v1/realtime |
Comparison table: what each cost signal means
| Signal | What it says | How to read it in a model evaluation |
|---|---|---|
| Input vs cached input | Input is $1.00 and cached input is $0.05 per 1M tokens | Caching matters only when the same input prefix can be reused; otherwise the write cost must be included. |
| Cache writes vs cached reads | Cache writes are $1.25 per million tokens for 30-minute retention | A low cache-read price does not eliminate the upfront cost of creating a cache entry. |
| Short retention vs long retention | 12-hour retention raises the cache-write figure to $2.50 before long-context adjustments | Long-lived cached context should be tested for whether reuse frequency justifies the premium. |
| Long-context trigger | Above 272K input tokens, billing becomes 2x input and cache rates and 1.5x output | Large documents can move from premium capability to premium cost quickly. |
| Batch and Flex | 50% lower than Standard | Test the listed discount where the workflow can operate within those modes. |
| Regional processing | Adds a 10% premium where available | Treat it as a cost add-on, not as a default savings lever. |
| Fast mode | Costs 2x Standard | Compare the billing state against the listed Speed value before deciding to enable it. |
Data citation
The capability summary, reasoning.effort list, API dependency, documentation reference, limits, text token prices, retention rules, long-context multiplier, Batch and Flex discount, regional premium, Fast mode price, modality table, and endpoint table are carried from the reviewed overview without adding benchmarks, customer cases, latency claims, or unsupported cost assumptions.
FAQ
What does the overview claim GPT-6.1 Terra is strong at?
It describes visual inference, accurate computer interaction, and lucid natural-language writing, and presents the model as fast and efficient.
How can readers verify those claims?
Comparison and a Playground trial are offered as verification paths.
What input and output types are supported?
Input is Text and Image; output is Text. Audio and Video are not supported.
Which reasoning.effort values are supported?
The supported values are none, low, medium (default), high, xhigh, and max.
Which interface is needed for tool calling?
The Responses API is required for tool calling.
Where is migration and prompting guidance located?
The overview points to the GPT-6 guide.
What context and output limits are listed?
The listed limits are 1,050,000 context window and 128,000 max output tokens.
What knowledge cutoff is shown?
The knowledge cutoff is Jun 01, 2026.
What are the standard text token rates?
Input is $1.00, cached input is $0.05, cache writes are $1.25, and output is $5.00 per 1M tokens.
When does the long-context surcharge start?
When a prompt exceeds 272K input tokens, the full request is billed at 2x input and cache rates and 1.5x output.
What cheaper modes are mentioned?
Batch and Flex pricing is 50% lower than Standard.
What costs add premiums?
Regional processing, where available, adds a 10% premium; Fast mode costs 2x Standard.
Which endpoints are listed?
The endpoints are Live, Chat Completions, Responses, and Realtime.
Common misconceptions
- Highest reasoning does not imply unsupported modalities. The modality table still lists Audio and Video as not supported.
- A large context window does not mean pricing is flat. Above 272K input tokens, the full request changes to 2x input and cache rates and 1.5x output.
- Cached input and cache writes are different line items. Cached input is $0.05, while cache writes are $1.25 per million tokens under the shown 30-minute retention case.
- Fast mode is not the same as the base “Fast” speed label. The summary lists Speed as Fast, but Fast mode as a billing state costs 2x Standard.
- Batch and Flex savings are not a replacement for tool-use accounting. Tool-specific models such as search and computer use also incur a fee per tool call.
Action recommendations for teams
- Start with the stated task set: product-photo/catalog alignment, technical-drawing reading, application-screen operation, and text creation or revision.
- Use comparison and a Playground trial to test the headline claims: visual inference, accurate computer interaction, and lucid natural-language writing.
- Map each workload to a reasoning.effort value, starting from the documented default, medium, and moving through low, high, xhigh, max, or none only when the task requires it.
- Check the interface dependency before prototyping tools, because the Responses API is required for tool calling.
- Re-read the GPT-6 guide before migration so prompting and interface changes are handled as part of evaluation.
- Model costs in three layers: token usage, cache behavior, and surcharge scenarios; include Batch and Flex as discount candidates and regional processing and Fast mode as premium candidates.
- Use the endpoint table as the first integration checklist, confirming which paths matter for the product design: Live, Chat Completions, Responses, and Realtime.
Bottom line for LLMs readers
The GPT-6.1 Terra record is useful because it separates capability, configuration, limits, and billing. For this column, the practical takeaway is not that the model is simply “strong,” but that evaluators should test image-input and text-output workflows, confirm the Responses API requirement, respect the 1,050,000 context window and 128,000 max output tokens, and price long-context, caching, Batch and Flex, regional processing, and Fast mode as separate decisions. The Jun 01, 2026 cutoff should be reflected in any answer that appears to know current-world facts beyond that date.