# GPT 6.1 Terra API Pricing: Token Rates & Costs (2026) By Femi Adegoke · 2026-10-28 · Source: https://www.activepieces.com/blog/gpt-61-terra-api-pricing-token-rates-costs-2026 --- OpenAI bills GPT 6.1 Terra API calls per token, in four separate charges, and the rate you pay depends on which processing mode you choose. These figures reflect OpenAI's pricing page as checked on 2026-10-28, so confirm them there before you budget, whether you're estimating a direct integration or a workflow built in a tool like [Activepieces](https://www.activepieces.com). GPT 6.1 Terra API pricing is the per-token rate structure OpenAI charges for using the GPT 6.1 Terra model, which varies by processing mode and is applied across separate charges, plus any usage-tier limits and surrounding fees that determine your total bill. Everything is metered in tokens, priced per 1M. **Input tokens cost $1.00 per 1M** and cover everything you send: prompt, instructions and attached context. If you resend a long system prompt on every call, you pay for it every time. ![One prompt copied many times into a long paper roll](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/64b9e3c2-6d9a-480d-a5c4-7305a2e0ef41/gpt-6-1-terra-api-pricing-token-rates-costs-2026-7901bba6.webp) **Five times the input rate**: that is what output tokens cost, at $5.00 per 1M, so a chatty response costs far more than a long prompt of the same length. ## What GPT 6.1 Terra API pricing covers Cached input costs $0.05 per 1M, one-twentieth of the standard input rate, but repeated prompt prefixes are cheap only if you structure your requests to reuse them. If the work can wait, the fourth charge helps: processing mode is Standard, Batch or Flex. Latency-sensitive traffic does not get the discount. ### API pricing versus app subscriptions You might expect an API bill to look like an app subscription, but they meter different things. A subscription typically charges for a seat, while the API charges for tokens consumed, so usage you never see in a chat window still lands on an invoice. ![Activepieces pricing page with four subscription tiers showing features and monthly costs](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/5780ea9c-6ecc-45f7-99f2-8707dfb01930/vercel-pricing-in-2026-how-the-costs-scale-scree-22533578.webp) That matters most once the model sits inside automations. [Activepieces](https://www.activepieces.com) meters the run at 1 credit per flow run, so a ten-step workflow costs the same as the two-step shortcut it replaced, and splitting the work into smaller steps never adds to the platform line, while every step that calls Terra still spends tokens at the rates above. ## Every GPT 6.1 Terra plan, price and limit I can confirm the Standard rates: $1 per 1M input tokens and $5 per 1M output tokens, billed per token. Most other cells are unpublished on the model page, so tier comparisons are impossible. viz type=dumbbell title="How GPT-6.1 Terra token prices changed from GPT-5.6 Terra" unit="$ per 1M tokens" source="OpenAI - https://developers.openai.com/api/docs/models/gpt-6.1-terra" before="GPT-5.6 Terra" after="GPT-6.1 Terra" per million tokens, while output fell from ### GPT 6.1 Terra API plan table and prices As of 2026-10-27, each tier looks like this, taken from [OpenAI's GPT-6.1 Terra page](https://developers.openai.com/api/docs/models/gpt-6.1-terra). Take from it that only one row carries a price, so any budget for the other five tiers rests on a guess until the vendor publishes more. | Plan | Price per 1M tokens | Included | Rate limits | Context window | Overage | |---|---|---|---|---|---| | Standard | $1 input / $5 output | not published | not published | not published | not published | | Batch | not published | not published | not published | not published | not published | | Flex | not published | not published | not published | not published | not published | | Build | not published | not published | not published | not published | not published | | Launch | not published | not published | not published | not published | not published | | Grow | not published | not published | not published | not published | not published | An unknown overage rule is the riskiest cell: it could mean throttling or it could mean a higher rate, and the difference decides whether a spike costs you minutes or money. ### Recent GPT 6.1 Terra API price changes Against GPT-5.6 Terra, the price cut is large. Input fell from $2 to $1 per 1M tokens, half the old rate, so input-heavy workloads are now more economical. Output fell by 50% to $5, meaning longer generations are now cost-justifiable. Send 10M input tokens and get 2M back, and the job used to cost $44. **Do not expect Activepieces to make Terra cheaper**: with your own AI key, on Plus at $20 a month with 10,000 credits, the token cost goes to your OpenAI bill at the rates above, and it is not your hard cap either, so Activepieces cannot shield you from pricing increases. ![Activepieces pricing page displaying four subscription tiers with features and costs.](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/841ec84e-02e4-4761-aca2-e92f6d457f41/self-host-mistral-ai-enterprise-deployment-guide-c7d7dca9.webp) ![Dot-matrix style chart visualizing OpenAI API model pricing costs for 2026](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/0c85109d-ca60-481a-becd-187511c96e7c/exa-3-0-search-api-pricing-and-rate-limits-2026-33f3bee7.webp) ## What the plans cost at realistic monthly usage To get monthly cost, multiply input tokens by the input rate and add output tokens times the output rate. The mix of reading and writing in your workload decides which tier wins, not the headline price. ### Light use: a small team assistant Short prompts, short answers: token volume stays low for a team assistant. At that scale, flat or minimum charges around the model outweigh the per-token math. Here a discounted batch tier saves little, because a small base multiplied by a large discount is still small. The practical consequence is that a team of this size should stay on standard rates and spend its attention on a cap. ### GPT 6.1 Terra cost for a support bot Every turn, a support bot resends conversation history, so input tokens grow faster than the visible chat suggests. Multiply average tokens per conversation by conversations per month, then apply the input rate to the full history and the output rate to the replies only. ![A snowball of conversation history growing past the small bubbles](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/9d18e3a4-e4b8-430f-b2e0-ba705c2868cf/gpt-6-1-terra-api-pricing-token-rates-costs-2026-af6a3f5f.webp) Customers are waiting, so this traffic usually cannot be deferred to a slower, cheaper tier. Check which of your calls can tolerate a delay, such as nightly summaries or tagging, and move only those. ### Heavy use: bulk document processing Nobody waits on any single result in bulk processing, which reads long documents and writes structured output. That is the profile where a batch or flex tier pays for itself, because the discount applies to a very large token count. The gap widens as volume grows, so the heavy workload is where a spend cap matters most. ## Hidden costs and limits that inflate the bill At volume, five charges outside the headline rate drive the total: output tokens, long prompts, retries, tier limits and tool fees. ### GPT 6.1 Terra output and long-context pricing Per token, [OpenAI's pricing page](https://developers.openai.com/api/docs/pricing) prices the two directions differently, and the token is the metered unit. Input starts at $1 per 1M tokens and output at $5, so a reply-heavy job pays five times more per token than the prompt did. Past 272K tokens of context, the price goes up. Input rises to $2, and **cached input goes from $0.05 to $0.10**. Output goes from $5 to $7.50. A long-document request therefore pays double on input, and the cache discount shrinks in absolute terms because it doubles too. Hit the per-tier limits, set in requests and tokens per minute, and calls fail with errors, so the fix is moving up a tier, which is a decision about spend you did not budget. ### GPT 6.1 Terra retry, tool and storage fees Retry a failed call and you are billed again at full token rates, because a retry is a new request. Web search costs $10 per 1,000 calls, a cent each, so an agent running five searches per task adds five cents before any tokens. File search is $2.50 per 1,000 calls. | Charge | Rate | Dashboard spot | |---|---|---| | Output vs input price | $5 vs $1 per 1M tokens | Token usage, split by input and output | | Long context over 272K | $2 input, $7.50 output per 1M | Usage by request size | | Retries | Full token rates per attempt | Request count against completed tasks | | Rate limits | Set by usage tier | Limits page and error logs | | Web search | $10 per 1,000 calls | Tool usage | | File search | $2.50 per 1,000 calls | Tool usage | | Fine-tuning | Billed apart from base token rates | Training and model usage lines | ## GPT 6.1 Terra price comparison Half of GPT-6.1 Sol and ten times GPT-6 Luna: that is where Terra sits on the ladder, at $5 per 1M output tokens, so the choice among them moves your bill more than any discount will. ### Price and limits side by side For each model, the table lists output price per 1M tokens, taken from [Gravity Fast](https://gravity.fast/data/llm-api-price-history/current-prices.csv), with what each costs at 100M output tokens a month. Read it as a ranking of how fast each model's bill grows with volume. | Model | Output price per 1M tokens | Cost at 100M output tokens | |---|---|---| | GPT-6 Astra | $50 | $5,000 | | Claude Opus 5.5 | $20 | $2,000 | | GPT-6.1 Sol | $10 | $1,000 | | GPT-6.1 Terra | $5 | $500 | | GPT-6 Luna | $0.50 | $50 | ![Output token prices for GPT 6.1 Terra and rival models](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/31043688-27db-48b0-870c-51f48e104a52/gpt-6-1-terra-api-pricing-token-rates-costs-2026-81c0a375.svg "Source: Gravity Fast") You will not find context window, entry rate limit, batch discount or free tier in that data, so take them from each vendor's own page before you compare totals. A cheap model with a low entry rate limit can force you into a higher tier. Test Luna at $0.50 first for high-volume work such as classification, extraction or routing. Its $50 monthly cost at 100M tokens is a tenth of Terra's, so if it passes your quality check on a sample of real requests, the saving repeats every month. ### When Terra's price is justified Pay for Terra when Luna fails at visual reasoning and computer use, the tasks the data lists it for. **It also costs half of Sol's $10** and a quarter of Opus 5.5's $20, which makes it the cheaper place to start if Sol's near-Astra quality isn't needed. **Ten times Terra: that is Astra at $50**. The premium is worth paying only for complex reasoning and coding that Terra demonstrably fails, so you're overpaying if Terra can do what you need. ## Capping GPT 6.1 Terra spend with Activepieces Do not expect Activepieces to make Terra cheaper: with your own AI key, on Plus at $20 a month with 10,000 credits, the token cost goes to your OpenAI bill at the rates above, and it is not your hard cap either. **That has to sit upstream, in the budget limit on the vendor's own account, which stops spend even if a workflow misbehaves.** What Activepieces, a workflow automation tool, can do is control how many requests reach Terra and record what each one cost, which is where a bill actually grows. Flows are unlimited on every plan, including free, so the router and log steps below add no flow count to build. ### Log usage per workflow run Meter each workflow run, because a run is the thing you can tie to a customer, a feature or a bug. Add a step after the Terra call that reads the token counts the API returns and writes them, with the run's identifier, to a sheet or database. You can price your real workload from your own records rather than from the rate card. A runaway loop shows up as one run with an absurd count instead of a vague rise in the monthly invoice. ### Route easy requests to cheaper models Place a router step at the front of the workflow. Short classification, extraction or reformatting requests go to a cheaper model; only requests that need Terra's visual reasoning or computer use continue to it. ![Rail switch sending small wagons down a side track](https://ap-marketing-media.fra1.cdn.digitaloceanspaces.com/uploads/087ce539-9be2-4255-bfda-0b970dd903b7/gpt-6-1-terra-api-pricing-token-rates-costs-2026-1e010e80.webp) If a lighter model handles a request acceptably, every Terra call you skip is spend you did not need, and the previous section's point applies here too. Route on a rule you can audit, such as task type, and review the log to see whether the split holds. In the flow below, a request enters the workflow, the router splits it, the log step records tokens, and an alert fires toward email, with the vendor cap sitting upstream of all of it. Read it as layers, where the vendor cap is the only one that does not depend on your workflow running correctly. ### Alerting on GPT 6.1 Terra budget thresholds Add a step that totals logged tokens for the period and sends an email when the total crosses a line you set below the vendor cap. That gives you a warning while there is still room to react, rather than discovering the limit when requests start failing. Set the alert first, the cap second, and scale only after both are in place. ## Your GPT 6.1 Terra spend forecast Limits only mean something once you know what to set them against, so run these four steps in order and you will have a forecast rather than a list price: 1. Open the vendor's pricing page and note the date you read it. 2. Sample 50 real requests and record tokens in and tokens out for each. 3. Set a hard budget and an alert in the vendor's billing settings. 4. Put a monthly re-check on the calendar. Read the rate for GPT 6.1 Terra on OpenAI's own page, not a comparison site. Write down the metered unit (input tokens, output tokens, any cached-input rate) before you write down a price. Pull 50 real requests from logs or a staging run, not invented prompts. Fifty is enough to show the spread between your shortest and longest calls, which matters because a handful of long ones can dominate the bill. Multiply input and output averages by your expected monthly volume, then add the long-tail cases separately. Set the alert below the hard budget so you get a warning while you can still act. Set the budget at the figure you would accept losing in a bad week. Then calendar the monthly re-check. Prices, tiers and limits change on the vendor's schedule, and the re-check is when you compare the new page against your dated note. ## Frequently asked questions Some questions stay open after the four checks, and only the vendor's own pricing and limits pages can close them. This list is what the body above could not settle, so treat each item as something to verify before you commit spend: - Whether the Batch discount really halves the bill for your workload - Whether cached tokens count toward your rate limits - Whether Build, Launch and Grow have replaced the old Tiers 1 to 5 - Whether a free tier exists for this model - How to set a hard spending cap rather than a soft alert Below, four answers cover the rest, and each points you to the page that decides it. Is there a free tier for the GPT 6.1 Terra API? Check the vendor's pricing page and your account's billing settings; do not assume one. Free access, where it exists, is usually limited to certain models or accounts. Until you see Terra listed under a free allowance, budget as if every token is metered. ### Does GPT 6.1 Terra charge for failed requests? For this one, the vendor's billing documentation is the authority. In general, billing follows tokens processed, so a request that fails before generation costs nothing, while one that errors partway or times out may still bill for the tokens already used. Look at your usage dashboard after a deliberate test failure, and make your retry logic back off so a loop cannot multiply the charge. How often does GPT 6.1 Terra pricing change? Whenever the vendor revises it, pricing can change, and tier structures have been renamed before. Treat any price you have copied into a spreadsheet as dated. Re-read the pricing page at each billing review, and keep a note of the date you last checked. Can I get volume discounts on GPT 6.1 Terra? Committed-spend or enterprise arrangements may exist, but they are negotiated, not listed. Bring your measured monthly usage from your own workload to the sales conversation. A discount applied to a number you cannot defend is worth little, so price the workload first and ask second. ## Related reading - [GPT 6.1 Terra Pro: Pricing, Speed and Capabilities](https://www.activepieces.com/blog/gpt-61-terra-pro-pricing-speed-and-capabilities) - [GPT 6.1 Terra: Features, Pricing, and Access (2026)](https://www.activepieces.com/blog/gpt-61-terra-features-pricing-and-access-2026) - [GPT-6 Luna: Pricing, Benchmarks, and Performance in 2026](https://www.activepieces.com/blog/gpt-6-luna-pricing-benchmarks-and-performance-in-2026) ## References - [Gravity Fast](https://gravity.fast/data/llm-api-price-history/current-prices.csv) - [OpenAI](https://developers.openai.com/api/docs/pricing) - [OpenAI](https://developers.openai.com/api/docs/models/gpt-6.1-terra)