# GPT-6.1 Terra: Pricing, Benchmarks, Cost per Task > GPT-6.1 Terra costs $1/$5 per million tokens, half of GPT-6.1 Sol. It scores 48 on Artificial Analysis for $0.40 a task, 56% of Sol's cost per task. Source: https://capitalandcompute.net/blog/gpt-6-1-terra-pricing-benchmarks/ Published: 2026-10-27 Updated: 2026-10-30 OpenAI released GPT-6.1 Terra on October 26, 2026, at $1 per million input tokens and $5 per million output. That is exactly half of GPT-6.1 Sol on both lines, and it replaces GPT-5.6 Terra, which charged $2 and $12\. The short verdict: for agent, computer-use and everyday API work on OpenAI, Terra is the new default to test, because the independent Artificial Analysis index runs a task on it for $0.40 against $0.72 on Sol for four fewer points (48 against 52). Keep Sol for the hardest coding jobs. Stay on Claude Haiku 5.5 or GPT-6 Luna for cheap bulk text, where a task costs a fraction of Terra’s. The half price does not halve the bill. Terra thinks for longer. | Model | Input / Mtok | Cached input | Output / Mtok | Where you can use it | | ------------------- | ------------ | ------------ | ------------- | -------------------------------------------------------------- | | **GPT-6.1 Terra** | **$1** | **$0.05** | **$5** | API (gpt-6.1-terra), ChatGPT Work and Codex on paid plans | | GPT-6.1 Sol | $2 | $0.10 | $10 | API (gpt-6.1-sol) | | GPT-5.6 Terra (old) | $2 | $0.20 | $12 | API, for workloads pinned to it | | GPT-6 Luna | $0.10 | $0.01 | $0.50 | API | | Claude Haiku 5.5 | $0.10 | $0.01 | $0.50 | Claude API, prompts up to 100,000 tokens ($0.50 / $2.50 above) | | Kimi K3.1 Pro | $2 | $0.20 | $10 | Kimi API | Rates are standard list prices read from each provider’s documentation on October 27, 2026\. OpenAI lists no introductory price or expiry for Terra, so the $1 and $5 is the rate to plan on. ## How much does GPT-6.1 Terra cost? GPT-6.1 Terra costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, per OpenAI’s [GPT-6.1 Terra model documentation](https://developers.openai.com/api/docs/models/gpt-6.1-terra). Batch and Flex processing bill half that. Prompts over 272,000 tokens bill the whole request at 2x input and 1.5x output. Three smaller lines matter if you build agents on it. Cache writes bill $1.25 per million tokens for 30-minute retention and $2.50 for 12-hour retention. Fast mode doubles everything. Regional processing, where offered, adds 10%. The cached read is the line to watch. At $0.05 it is 95% off fresh input, and an agent that re-reads a long system prompt and file context on every turn pays that rate for most of its input. GPT-6.1 Sol charges $0.10 for the same read and GPT-5.6 Terra charged $0.20, so a cache-heavy loop moving from the old Terra sees its biggest saving there, a 75% cut, larger than the headline 50%. Against the model it replaces, the cut is not uniform either. Input falls 50%, from $2 to $1\. Output falls 58%, from $12 to $5\. Output dominates most agent bills, so “50% cheaper” (OpenAI’s framing, repeated in [BleepingComputer’s launch report](https://www.bleepingcomputer.com/news/artificial-intelligence/openai-launches-gpt-61-terra-with-3x-faster-ai-agents-50-percent-lower-pricing/)) undersells the move from GPT-5.6 Terra. ## What does a task actually cost on GPT-6.1 Terra? Per token, Terra is half of Sol. Per task, it is a little over half, because it uses about twice the tokens to reach its answer. [Artificial Analysis](https://artificialanalysis.ai/models/gpt-6-1-terra), the independent benchmarking firm, ran its Intelligence Index on Terra at max effort and measured $0.40 per index task, with 130 million output tokens spent across the run. The same run on [GPT-6.1 Sol](https://artificialanalysis.ai/models/gpt-6-1-sol) cost $0.72 a task and used 67 million output tokens. Halve the rate, nearly double the tokens, and the task lands at 56% of Sol’s cost. For that, Terra scores 48 where Sol scores 52. __GPT-6.1 Terra relative to GPT-6.1 Sol: rate, tokens, cost and score__ | Metric | GPT-6.1 Terra as a multiple of GPT-6.1 Sol | Values | | --------------------------- | ------------------------------------------ | ------------------ | | Input price per token | 0.50x | $1 vs $2 per Mtok | | Output price per token | 0.50x | $5 vs $10 per Mtok | | Output tokens per index run | 1.94x | 130M vs 67M | | Cost per index task | 0.56x | $0.40 vs $0.72 | | Intelligence Index score | 0.92x | 48 vs 52 | GPT-6.1 Terra as a multiple of GPT-6.1 Sol on a log axis, so 0.5x and 2x sit the same distance from Sol. Half the rate per token, but nearly twice the output tokens, so a task costs 0.56x as much for a 0.92x score.Source: [Rates: OpenAI model documentation. Tokens, cost per task and index score: Artificial Analysis Intelligence Index, max effort, read October 27, 2026](https://artificialanalysis.ai/models/gpt-6-1-terra) Here’s how that works. A model’s bill for a task is its rate multiplied by the tokens it spends, and a reasoning model spends most of its tokens thinking before it answers. Terra is a smaller, cheaper model per token, and on hard questions it compensates by thinking longer. So the price sheet and the invoice disagree. The rate card says 50% off. The task says 44% off. That is still a large saving. And against its own predecessor the gap is far wider: Artificial Analysis puts [GPT-5.6 Terra at $1.40 a task](https://artificialanalysis.ai/articles/gpt-6-1-terra-pushes-the-cost-efficiency-frontier) at max effort, so the new Terra scores 6 points more for under a third of the cost. | Model at max effort | AA Intelligence Index | Cost per index task | Output tokens on the run | | ------------------- | --------------------- | ------------------- | ------------------------ | | Claude Sonnet 5.5 | 56 | $5.46 | 420M | | Kimi K3.1 Pro | 55 | $2.11 | not read | | GPT-6.1 Sol | 52 | $0.72 | 67M | | **GPT-6.1 Terra** | **48** | **$0.40** | **130M** | | Claude Haiku 5.5 | 43 | $0.21 | 440M | | GPT-5.6 Terra | 42 | $1.40 | not read | | GPT-6 Luna | 38 | $0.07 | 140M | Artificial Analysis values read October 27, 2026 from each model’s page, except Kimi K3.1 Pro (read October 24) and GPT-5.6 Terra (from Artificial Analysis’s launch article, since the model has left its leaderboard). Kimi K3.1 Pro is listed without an effort label. Max effort is not the only setting worth running. Artificial Analysis’s launch article reports that at low and medium effort Terra gains 7 to 13 index points over GPT-5.6 Terra, and that its xhigh and low settings sit on the cost-efficiency frontier of its Coding Agent Index, where Terra scores 48 against Sol’s 54 and Luna’s 31\. If max effort’s long thinking makes a job slow, low effort is the first setting to try, not a downgrade to avoid. ## Where can you use GPT-6.1 Terra? In three places, all live on launch day. Developers call it as `gpt-6.1-terra` on the Chat Completions, Responses and Batch endpoints. ChatGPT Plus, Pro, Enterprise and Edu users get it in ChatGPT Work and in Codex, as reported by BleepingComputer. The specification matches the rest of the GPT-6.1 line: a 1,050,000-token context window (922,000 tokens of input, 128,000 of output), text and image in, text out, six reasoning-effort levels from none to max with medium as the default, and a knowledge cutoff of June 1, 2026, a month later than Sol’s. It has no Realtime, fine-tuning or embeddings endpoint. GPT-5.6 Terra stays on OpenAI’s price list at $2 and $12\. Anything pinned to it keeps paying the old rate until someone changes the model string. ## What did OpenAI claim, and what holds up? OpenAI pitches Terra at computer use. Its [launch post on X](https://x.com/OpenAI/status/2114780711664398349) says the model “understands what’s on screen, takes action across apps, writes naturally.” Its launch numbers, as reported by BleepingComputer, are OpenAI’s own: * **Terminal-Bench 4.0, max effort:** 50.3%, against 21.5% for GPT-5.6 Terra. * **Internal error rate:** down from 53.6% to 16.7% at low effort, and to 4.3% at max effort, which OpenAI says nearly matches GPT-6 Astra and GPT-6.1 Sol. * **Speed:** tasks finished more than three times faster than Astra and Sol. * **Agents’ Last Exam:** 15 points above GPT-5.6 Terra at medium effort, for about a third of the cost. The internal error test is OpenAI’s and can’t be checked. The Terminal-Bench figure can be, eventually: the [independent Terminal-Bench board](https://www.tbench.ai/) publishes scores and run costs, and it records GPT-6.1 Sol and GPT-6 Astra tied at 58.18%, at $3.30 and $17.02 per solved task. No Terra entry could be found on it on October 27\. Artificial Analysis’s own run points the same direction as OpenAI’s number, a gain of about 20 points that leaves Terra roughly level with Sol at max effort. The “on par with GPT-6 Astra at a tenth of the per-token price” line is half true. The per-token part is exact: Astra lists $10 and $50, ten times Terra’s $1 and $5\. “On par” holds on OpenAI’s agent test, not on the broader independent index, where Astra scores 53 and Terra 48. One independent number is better than anything OpenAI claimed. On Artificial Analysis’s AA-Omniscience test, Terra’s hallucination rate is 38%, against 88% for GPT-5.6 Terra, a 50-point drop. The old Terra answered almost every question it didn’t know. The new one declines more often. Speed versus first token Artificial Analysis measures Terra at 90.9 output tokens per second at max effort, faster than Sol’s 69.4, but with a time to first token of about 132 seconds, because max effort thinks before it writes. OpenAI’s “three times faster” is a task-completion claim from its own tests; for interactive use, a lower effort setting is what makes it feel fast. ## Is GPT-6.1 Terra better than GPT-6.1 Sol? No, but it’s cheaper by more than the score gap. Sol scores 52 and Terra 48 on the Artificial Analysis index, and 54 against 48 on its Coding Agent Index. Sol costs 1.8 times as much per task. On Terminal-Bench 4.0 the two are close on both OpenAI’s figure and Artificial Analysis’s reading. So the split is about the work. For an agent that clicks through a web app, triages a ticket queue or drafts documents, the four index points rarely decide the outcome, and the 44% saving per task compounds across every run. Start on Terra and route the jobs it fails up to Sol. The context window, endpoints and surcharge rules are identical, so switching between them is a one-line model-string change. The detail on Sol’s own launch, and the GPT-6 tiers it replaced, is in [GPT-6 Sol and Luna: pricing and benchmarks](/blog/gpt-6-sol-luna-pricing-benchmarks/). ## How does GPT-6.1 Terra compare with Claude Haiku 5.5 and Sonnet 5.5? Terra sits between Anthropic’s two tiers, closer to Haiku on cost and closer to neither on score. Against [Claude Haiku 5.5](/blog/claude-haiku-5-5-pricing-benchmarks/), Terra scores 5 points higher (48 against 43) for about twice the cost per task ($0.40 against $0.21). Haiku’s rate card is a tenth of Terra’s, but Haiku spends 440 million output tokens on the index run, more than three times Terra’s 130 million, which closes most of the gap. If your work is short, high-volume text (classification, extraction, summaries), Haiku is still cheaper. If it needs multi-step reasoning, the extra five points are cheap. Against [Claude Sonnet 5.5](/blog/claude-sonnet-5-5-pricing-benchmarks/), Terra scores 8 points lower (48 against 56) for 7% of the cost per task ($0.40 against $5.46). Sonnet 5.5 is second on the whole index. Paying 13.7 times as much for it makes sense on the jobs where those 8 points are the difference between done and not done, and almost nowhere else. ## How does GPT-6.1 Terra compare with Kimi K3.1 Pro? Moonshot’s [Kimi K3.1 Pro](/blog/kimi-k3-1-pro-pricing-benchmarks/), released three days earlier on October 23, scores 55 on the same index at $2.11 a task, at a $2 and $10 rate card, the same as Sol’s. It is 7 points above Terra for 5.3 times the cost per task. It is also a Chinese model, with weights promised by November 4, which rules it out for some buyers and makes it self-hostable for others. For a team choosing on price per point alone, Terra is the cheap way to reach the high 40s, and Kimi K3.1 Pro is the cheapest way to reach 55, a score no OpenAI model reaches at any price. ## Which should you use? * **Agents, computer use and everyday API work on OpenAI:** GPT-6.1 Terra. Try low and xhigh effort before max. * **The hardest coding and long agent runs:** GPT-6.1 Sol, or Astra if the job justifies 10 times the rate. Route up from Terra on failure rather than starting there. * **Cheap high-volume text:** GPT-6 Luna at $0.07 a task, or Claude Haiku 5.5 at $0.21 if it scores better on your prompts. * **Near-frontier quality at the lowest cost:** Kimi K3.1 Pro at $2.11 a task, if a Chinese model is allowed. * **Still on GPT-5.6 Terra:** move. The new tier is cheaper on every line and scores higher on every independent measure read here. ## How do you switch to GPT-6.1 Terra? 1. Change the model string to `gpt-6.1-terra` in your API calls. Nothing else in the request needs to change if you already use GPT-6.1 Sol or GPT-5.6 Terra. 2. Leave reasoning effort at the default (medium) for a first run, then compare low and xhigh on a sample of real tasks. Measure cost per finished task, not per token. 3. Put stable context (system prompt, tool definitions, reference files) at the front of the prompt so it hits the $0.05 cached rate. 4. Keep prompts under 272,000 tokens where you can. Above that, the whole request bills at 2x input and 1.5x output. 5. Send work that isn’t time-sensitive through Batch or Flex for half price. Current rates for every model on this site, each dated and sourced, are on the [AI model tracker](/ai-models/), and every October release is in [new AI models released in October 2026](/blog/new-ai-models-october-2026/). The [coding agent cost-per-task tracker](/coding-agent-cost-per-task/) models what a task costs across harnesses. ## Frequently asked questions How much does GPT-6.1 Terra cost? GPT-6.1 Terra costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Batch and Flex are half price. Prompts over 272,000 tokens bill the whole request at 2x input and 1.5x output. OpenAI lists no introductory price, so this is the standard rate. Is GPT-6.1 Terra half the price of GPT-6.1 Sol? Per token, yes, exactly. Per task, not quite. Artificial Analysis measured $0.40 per index task on Terra against $0.72 on Sol at max effort, 56% of the cost, because Terra spent 130 million output tokens on the run against Sol’s 67 million. Terra scores 48 to Sol’s 52. When was GPT-6.1 Terra released? OpenAI released GPT-6.1 Terra on October 26, 2026, in the API as gpt-6.1-terra and in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. It replaces GPT-5.6 Terra, which stays on the price list at $2 and $12 per million tokens. Is GPT-6.1 Terra better than GPT-5.6 Terra? Yes, on every independent measure read here. Artificial Analysis scores it 48 against 42 at max effort, for $0.40 per index task against $1.40, and its hallucination rate on AA-Omniscience fell from 88% to 38%. OpenAI reports 50.3% on Terminal-Bench 4.0 against 21.5%. Which is cheaper per task, GPT-6.1 Terra or Claude Haiku 5.5? Claude Haiku 5.5, at $0.21 per Artificial Analysis index task against $0.40 for GPT-6.1 Terra. Terra scores 5 points higher, 48 against 43\. Haiku is the better pick for short, high-volume text work and Terra for multi-step reasoning and agents. ## Sources OpenAI (2026). _GPT-6.1 Terra_. OpenAI API model documentation: $1.00 input, $0.05 cached input, $5.00 output per Mtok; cache writes $1.25 (30-minute) and $2.50 (12-hour); 1,050,000-token context, 128,000 max output, knowledge cutoff June 1, 2026; effort levels none to max. Verified 2026-10-27. Artificial Analysis (2026). _GPT-6.1 Terra_ model page. Independent benchmark index: Intelligence Index 48 at max effort, $0.40 per index task, 130M output tokens, 90.9 output tokens per second. Verified 2026-10-27. Artificial Analysis (2026). _GPT-6.1 Terra pushes the cost efficiency frontier_. Launch analysis, October 26, 2026: +6 index points and $0.40 against $1.40 per task versus GPT-5.6 Terra, Coding Agent Index 48 (Sol 54, Luna 31), AA-Omniscience hallucination rate 38% against 88%. Verified 2026-10-27. Artificial Analysis (2026). Model pages for _GPT-6.1 Sol_, _Claude Sonnet 5.5_, _Claude Haiku 5.5_ and _GPT-6 Luna_. Intelligence Index, cost per index task and output tokens at max effort. , , , Verified 2026-10-27. BleepingComputer (2026). _OpenAI launches GPT-6.1 Terra with 3x faster AI agents, 50 percent lower pricing_. Secondary report of OpenAI’s vendor-reported launch figures. Accessed 2026-10-27. OpenAI (2026). _GPT-6.1 Terra_ launch post on X, October 26, 2026\. Accessed 2026-10-27. Terminal-Bench (2026). _Terminal-Bench 4.0 leaderboard_. GPT-6.1 Sol and GPT-6 Astra at 58.18%, $3.30 and $17.02 per solved task (read 2026-10-20); no GPT-6.1 Terra entry found on 2026-10-27\.