GPT-6.1 Terra is real and purchasable. It appears as a listed model on OpenAI's official API platform pricing page at $1.00 per 1M input tokens and $5.00 per 1M output tokens, with a 1.05M context length, 128K max output tokens, and a knowledge cut-off of June 1, 2026 [2]. The OpenAI developer community announcement confirms it is rolling out now to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and that developers can access it through the API under the model ID gpt-6.1-terra [4]. For enterprise SaaS teams, the practical question is no longer whether to wait for confirmation, but which workloads to move onto it and which to leave on Sol or Astra. The short answer based on the verified record: high-volume agent workloads involving screen understanding and cross-app action are the intended target, standard-priced token costs drop by half versus GPT-6.1 Sol and one-tenth versus GPT-6 Astra, but the headline price only applies below 272K context, and OpenAI itself steers deep research and complex coding elsewhere.
Verified release facts: two official sources, one caveat
The evidence trail is straightforward. The model pricing card on openai.com/api lists GPT-6.1 Terra alongside GPT-6 Astra and GPT-6.1 Sol with live pricing [2], and the official announcement post states availability and the API model ID directly [4]. A community thread from September 2026 asking why there was no GPT-6 Terra [3] confirms the version slot sat empty until late October, so this is a very fresh addition rather than a long-documented tier.
Two caveats belong in the record. First, the formal announcement page at openai.com/index/introducing-gpt-6-1-terra/ was not retrievable in this research pass; the release date can only be inferred as around October 26, 2026 from community thread timestamps [1]. Second, the API platform page's footer 'Latest Advancements' list still names only GPT-6.1 Sol and GPT-6 Astra, not Terra [5] — a minor internal inconsistency on OpenAI's own site. The pricing card is the direct evidence and controls, but procurement teams that verify via the marketing page may momentarily see the model missing.
Pricing: the hardest number in the announcement
The price structure is the clearest enterprise impact. Terra's standard API price of $1/$5 per 1M tokens is stated by OpenAI as 50% lower than GPT-5.6 Terra, and it is exactly half of GPT-6.1 Sol's $2/$10 [2][4]. Against GPT-6 Astra at $10/$50 [2], Terra costs one-tenth per token — and the announcement claims it matches Astra on the Agents' Last Exam agentic benchmark at that one-tenth price [4]. For a SaaS product running agents at volume, moving a workload from Sol to Terra cuts inference spend by half at standard rates; from Astra, by 90%.
The critical qualifier: OpenAI's pricing page states these figures "reflect standard processing rates for context lengths under 272K" [2]. A model with a 1.05M context window being sold at $1/$5 does not mean million-token inputs bill at $1. Pricing above 272K, batch-processing differentials, and regional-processing rates were not in the retrieved materials. Any cost model that assumes the headline rate across the full context range will be wrong until the model card's long-context tier is checked. For teams whose agent runs routinely fill large context windows, this is the single most important line item to close before migrating.
Context and specs: same window, newest knowledge cut-off in the family
Terra matches Sol and Astra on 1.05M context and 128K max output [2], so migrating agent architectures does not change the input/output envelope. One spec actually differs in Terra's favor: its knowledge cut-off of June 1, 2026 is newer than Sol's and Astra's April 30, 2026 [2]. Teams that built retrieval layers partly to compensate for stale model knowledge may find fewer refresh calls justified for Terra-routed workloads, though that is a per-workload measurement, not a blanket claim.
What the pricing page does not give is Terra's rate limits. The model card link in the announcement points to them [4], but the specific numbers were not retrievable here. High-concurrency SaaS deployments cannot confirm throughput ceilings until those figures are in hand, so load testing against real traffic patterns remains a prerequisite to any production cutover.
Capability boundary: fast and cheap agents, with OpenAI's own exclusion zone
The benchmark claims, all from OpenAI's announcement: Terra scores 50.3% on Terminal-Bench 4.0, up from 21.5% for GPT-5.6 Terra [4]; it substantially outperforms GPT-5.6 Terra on Agents' Last Exam, matches GPT-6 Astra at one-tenth the per-token price, and finishes tasks over 3x faster on average than both Astra and GPT-6.1 Sol [4]. The Agents' Last Exam figure is described in text only; the underlying chart data was not retrievable, so treat the magnitude as vendor-reported until independently confirmed.
Just as important is what OpenAI says Terra is not for. The announcement is explicit: "For in-depth research or complex coding, choose GPT-6.1 Sol or GPT-6 Astra," positioning Terra as a model that "understands what's on screen, takes action across apps" across the whole workday [4]. This draws a clean migration line for SaaS engineering teams. Moving a coding-assistant product from Sol down to Terra runs against the vendor's own positioning. Moving back-office agent pipelines — document reading, application-screen interpretation, executing actions across internal systems — onto Terra is exactly the workload it was built and priced for. The 3x average task-completion speed claim compounds the economics: faster completion means lower wall-clock latency and more tasks per unit of spend, assuming the quality bar holds on your specific tasks.
Integration path: a model-parameter change, not a re-platform
Terra introduces no new integration surface. The OpenAI API platform's agent stack — Agents SDK, Responses API, and built-in tools including web search, file search, and remote MCP — applies to models on the platform [2]. Enterprise-grade controls listed on the platform page include no training on customer data, zero data retention by request, HIPAA Business Associate Agreements, SOC 2 Type 2 compliance, and data residency controls [2].
One attribution limit matters here: those compliance features are documented as platform-level properties, not itemized per-model commitments in the retrieved materials. Enterprise customers who need Terra covered under their existing agreements should get written confirmation from their OpenAI account team before routing regulated workloads to it. Separately, for teams evaluating data residency, the documented regions cover US, EU, UK, JP, CA, KR, SG, IN, AU, and UAE [5] — mainland China is not among them, so cross-border data-flow compliance for China-operated services requires independent assessment on top of any model choice.
The decision, in full
The verified picture as of October 27, 2026: GPT-6.1 Terra is a shipping, API-accessible model at gpt-6.1-terra, half the standard price of GPT-6.1 Sol and one-tenth of Astra below 272K context, with the same 1.05M/128K envelope, the newest knowledge cut-off in the GPT-6.1 family, and vendor-reported agentic gains of 50.3% on Terminal-Bench 4.0 and 3x faster average task completion [2][4]. If your product runs high-throughput agent workloads built around screen understanding, cross-app action, or document processing, Terra is the tier OpenAI built for that use, and an A/B migration evaluation in a test environment is the rational next step. If your core product value is deep research or complex coding, Sol and Astra remain the vendor-designated choices and nothing in this release changes that. And before any cost model gets committed to a roadmap: the $1/$5 rate stops at 272K context, rate limits are not yet in hand, and platform compliance terms should be confirmed in writing to cover Terra specifically. Those three gaps determine whether the half-price math survives contact with your actual traffic.