GPT-5.6 Terra API Model Parameters and Tool Configuration

One-sentence takeaway

This official model page provides Terra's model ID, reasoning levels, context and output limits, pricing, and the Responses API tool surface, making it a practical configuration baseline before integration.

Use cases

• Suitable tasks: Text and code tasks that require a balance among intelligence, speed, and cost, as well as agents that need to call search, file, code interpreter, or MCP tools.

• Unsuitable tasks: Audio or video input/output and tasks requiring fine-tuning; the page does not promise a success rate for any specific business use case.

• Applicable model version: gpt-5.6-terra (the alias currently displayed on the page).

• Applicable client, agent, or API: OpenAI API, especially the Responses API; Chat Completions is also listed as a supported endpoint.

• Recommended reasoning level and parameters: Start with medium (the default); try none or low for short tasks where latency or cost is sensitive, and step up to high, xhigh, or max for complex debugging and long-horizon planning.

Ready-to-use content

The following is a minimal configuration snippet organized from the fields on the official page; it does not include authentication, SDK initialization, or a complete implementation of business tools.

model = "gpt-5.6-terra"
reasoning.effort = "medium"  # none | low | medium | high | xhigh | max
endpoint = "/v1/responses"

The available tools listed on the Responses API page include web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search. The page also marks structured outputs and function calling as Supported.

Testing/workflow steps

1. Fix gpt-5.6-terra and first run a set of representative tasks with medium, recording the input, output, reasoning level, number of tool calls, and total elapsed time.

2. For tasks that only require reading, rewriting, or small-scope fixes, run a small-sample A/B comparison with low and medium; raise the reasoning level only when the quality difference is clear.

3. Use the Responses API for tasks requiring multi-round tool coordination, and enable only the tools actually needed; keep large intermediate objects returned by tools on the tool side to avoid sending them back to the model unnecessarily.

4. When a single input approaches 272K tokens, calculate the full request cost first; the official rule is that once this threshold is exceeded, the entire request is priced at 2x the input rate and 1.5x the output rate, after which you can consider splitting documents or repositories into chunks if necessary.

5. Record whether caching is used: calculate cached reads, cached writes, and uncached input separately at the rates published on the page, and do not treat a cache hit as free.

6. If verification involves audio/video or fine-tuning, reject it at the routing layer first or switch to a model that supports the capability.

Original evidence and data

• Model positioning: The page describes Terra as a GPT-5.6 model that “balances intelligence and cost,” roughly corresponding to the mini tier of the earlier GPT-5 series.

• reasoning.effort: Supports none, low, medium (default), high, xhigh, and max.

• Context window: 1,050,000 tokens; maximum output: 128,000 tokens; knowledge cutoff date: 2026-02-16.

• Pricing: $2.00 per 1M input tokens, $0.20 per 1M cached input tokens, and $12.00 per 1M output tokens; cached writes are charged at 1.25 times the uncached input rate.

• Long-input rule: A complete request exceeding 272K input tokens is priced at 2x the input rate and 1.5x the output rate.

• Modalities: Text input/output and image input; audio and video are marked as unsupported.

• API capabilities: Streaming output, function calling, and structured outputs are supported; fine-tuning is unsupported.

• Responses API tools: Web search, file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, MCP, and tool search are all marked as supported.

• Tier 1 rate limits: 500 RPM, 500,000 TPM, and a 1,500,000 batch queue limit; use the values actually shown for the account for higher Tiers.

Limitations and boundaries

• This is a capability and pricing directory, not an independent quality evaluation; a specific business's accuracy cannot be inferred from “Default” or the price alone.

• The endpoints listed on the page do not mean that every endpoint has the same tool capabilities; the tool description is explicitly limited to the Responses API.

• The 272K threshold applies to the full request for billing, rather than adding a surcharge only to the excess; chunking should be validated in budget tests for long-document workflows to determine whether it is more cost-effective.

• The page displays an alias; to lock in behavior, confirm the specific snapshot available before launch. The current page does not disclose an independently selectable snapshot name.

Source excerpt or observation (short quotation for compliance only)

The official positioning statement is “GPT-5.6 model that balances intelligence and cost”. This supports the “balanced tier” positioning, but cannot replace a business benchmark.