GPT-5.6 Terra

gpt-5.6 terra is OpenAI's middle tier in the 5.6 line, built for teams that want frontier-grade reasoning without paying GPT-5.5 Pro rates on every call. It slots between the everyday GPT-5.5 and the precision-heavy GPT-5.5 Pro, and it's the model worth reaching for when a job needs real multi-step reasoning — refactoring a large codebase, walking through a long contract, chaining tool calls — but doesn't justify Pro pricing on every token. The trade-off is straightforward: you give up a little of Pro's ceiling in exchange for meaningfully lower cost and the same 1,050,000 token window.

What Terra Actually Does Well

Terra is a reasoning model, not a plain chat model, so it thinks through multi-step problems rather than pattern-matching a single-pass answer. That matters for agentic work: planning a sequence of tool calls, debugging across several files, or synthesizing a long research document into something usable. It supports function calling and vision, and takes text, image, and PDF input while returning text — a practical combination for teams building coding agents or document-review pipelines that need to reason over screenshots or scanned files alongside plain text.

Context Window and Price, in Real Numbers

The window is 1,050,000 tokens, with a 128,000 token cap on output. In practice that's enough room to load a mid-sized repository, a full legal filing, or a stack of PDFs into a single call and still leave headroom for the model's reasoning trace and response. Pricing sits at $2.00 per million input tokens and $12.00 per million output tokens, with cached input priced at $0.20 per million tokens — a 90% discount versus fresh input. For workloads that repeatedly re-send the same system prompt or reference document (agent loops, RAG pipelines, long chat threads), that cache price is the number to design around, not the headline input rate.

Where Terra Sits Against GPT-5.5 and GPT-5.5 Pro

OpenAI's other current-generation models bracket Terra on both sides. GPT-5.5 runs $5.00 in / $30.00 out per million tokens with $0.50 cached input — more expensive than Terra despite being positioned as the "default" tier. GPT-5.5 Pro runs $30.00 in / $180.00 out with no published cache discount at all, aimed squarely at slow, precision-heavy reasoning where cost is secondary. Terra undercuts both on price while matching their 1,050,000 token context window and 128,000 token output cap, which makes it the pragmatic default for high-volume reasoning work — reserve Pro for the specific tasks that actually need its extra ceiling.

The Honest Limitation

Terra only outputs text — no image generation, no audio. If a workflow needs the model to produce a diagram, an edited image, or spoken output, this isn't the model for that leg of the pipeline; pair it with a dedicated generation model instead. It's also worth testing your own document types against the 1,050,000 token window before committing — PDF parsing quality and how much of that window survives image-heavy inputs varies enough between vendors that it's worth a quick pilot before routing production traffic through it.

Practical Fit

• Context window: 1,050,000 tokens, output capped at 128,000 tokens
• Input: $2.00 per million tokens; output: $12.00 per million tokens
• Cached input: $0.20 per million tokens, a 90% discount on repeated context
• Reasoning, function calling, and vision are all supported
• Input modalities: text, image, PDF; output: text only
• Not open-weight — API access through OpenAI only

Terra is the model to reach for when a task genuinely needs step-by-step reasoning over a large document or a multi-tool agent loop, and the budget doesn't stretch to Pro-tier pricing on every request.