GPT-6.1 Terra API Pricing: $1/$5 — Half of Sol61, a Tenth of Astra

Short answer: OpenAI launched GPT-6.1 Terra on October 26, 2026, and the API rate card is the cleanest one in the lineup: $1.00 input, $5.00 output, $0.05 cached input per million tokens on the standard tier. That is exactly half of GPT-6.1 Sol's $2/$10 on every single line — input, output, cache reads, cache writes — and exactly one-tenth of Astra's $10/$50. If your workload runs on Sol61 today and Terra can do the job, your bill halves overnight. The rest of this article is the receipt.

Published October 27, 2026 · Pricing verified October 27, 2026

What OpenAI announced

Terra slots into the middle of the lineup — Luna, Terra, Sol, Astra — as the model for work that happens on a screen: visual reasoning, precise computer use, and what OpenAI calls "clearer, more natural writing." Think matching product photos to catalog entries, reading technical drawings, acting on application screens, drafting and revising text. It shipped in the API as gpt-6.1-terra and in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users. OpenAI says standard prices are 50% lower than its predecessor, GPT-5.6 Terra. The current family table, as listed on OpenAI's pricing page, now reads:

Same frame as the rest of the family: a 1,050,000-token context window, 128K max output tokens, and a June 1, 2026 knowledge cutoff. Two details worth knowing before you route traffic: reasoning.effort supports none, low, medium (default), high, xhigh, and max — and tool calling requires the Responses API, not Chat Completions. Prices verified October 27, 2026 — always confirm on the official pricing page.

On capability, I will only relay OpenAI's own claims, because this site measures pricing, not benchmarks: Terra matches Astra on its Agents' Last Exam at one-tenth the token price, finishes tasks over 3× faster than Astra and Sol61 on OpenAI's timing runs, and scores 50.3% on Terminal-Bench 4.0 at maximum effort versus 21.5% for its predecessor. OpenAI's own guidance is also blunt about the ceiling: "For in-depth research or complex coding, choose GPT-6.1 Sol or GPT-6 Astra." Test your own workload before moving production traffic — the price is certain, the capability claim is theirs.

What one request actually costs

Take a computer-use turn: 50K input tokens at 80% cached, 5K output, standard tier. The arithmetic is short enough to check by hand:

• Terra: 10K fresh input × $1 + 40K cached × $0.05 + 5K output × $5, all per million = $0.01 + $0.002 + $0.025 = $0.037 per request.
• Sol61: the same turn bills $0.074 — exactly double, as it must be, since every rate is double.
• Astra: $0.39 per request — Terra runs the same turn for under a tenth of that.
• Luna: $0.0039 — still roughly a tenth of Terra, for workloads simple enough to live there.

Scale that turn to 100K requests a month and the gap stops being a rounding error: Terra bills $3,700/month, Sol61 $7,400/month, Astra $39,000/month, Luna $390/month. Or strip it down to the unit the whole industry quotes — 1M input tokens plus 1M output tokens, no caching: Terra $6, Sol61 $12, Astra $60, Luna $0.60. The practical rule falls out of the table: if you already have a Sol61 estimate for a workload, halve it — that is your Terra estimate, on standard, long-context, and every processing mode alike, because the halving holds on all five rate columns.

Where Terra wins — and where it doesn't

The honest way to place Terra is against its two neighbors, not against Astra. Against Luna, Terra costs 10× per token — so anything Luna already handles well (classification, extraction, routing) should stay on Luna, and moving it to Terra would be paying ten times more for capability you don't use. Against Sol61, Terra costs half — so the question is only whether your task survives the step down. OpenAI's positioning points at the workloads where it should: screen work, forms, invoices, drawings, drafting. The invoice example in the launch post is telling — high-volume, repetitive, visual, and exactly the kind of job where a 50% rate cut compounds across thousands of requests a day. What Terra is not, by OpenAI's own admission, is the pick for the hardest coding or research; that stays Sol61 and Astra, and no rate cut changes that.

One more wrinkle for agent builders: reasoning tokens are billed as output, and Terra's effort dial runs from none to max. Cranking effort to max on a cheap model can quietly eat the savings the rate card just gave you — the output line is where agent bills actually live. If you switch, watch output tokens per task first, not input.

Long context, cache writes and processing modes

Above 272,000 input tokens, the entire request reprices: $2 input, $0.10 cached read, $7.50 output per 1M tokens — 2× input and cache rates, 1.5× output, applied to the full request, not just the excess. Exactly 272,000 stays short context. Cache writes are their own line: $1.25 per 1M for 30-minute retention, $2.50 for 12-hour, before long-context adjustments. Standard costs 1×; Fast 2×; Flex and Batch 0.5×; EU regional processing adds 10%.

Example: 300K input, 10K output, no caching costs $0.675 per request on Standard ($0.60 input + $0.075 output), $1.35 on Fast, $0.3375 on Flex or Batch, and $0.7425 on EU Standard. The 272K cliff matters more for Terra than for Luna, because Terra's whole pitch is long, screen-heavy sessions — the kind that drift across the line mid-conversation. Compact before it, or pay double on input from that turn on.

Calculate your workload

Terra is so new that our calculator doesn't list it yet — the rates above come straight from OpenAI's pricing page, verified October 27, 2026. Until it lands in the calculator, the shortcut is exact, not approximate: run your workload on GPT-6.1 Sol and halve the result. For the broader comparison, see GPT-6.1 Sol's pricing and the GPT-6 vs Claude comparison.

Official sources, verified October 27, 2026: GPT-6.1 Terra model documentation, OpenAI API pricing, and OpenAI's GPT-6.1 Terra announcement (October 26, 2026).