AI
“OpenAI launched GPT-6.1 Terra with 50% lower API prices and 3x faster computer use than previous models.”
Plain restatementOpenAI has released a model called GPT-6.1 Terra. Its API prices are 50% lower than some comparison model, and it performs computer-use tasks about three times faster than earlier models.
OpenAI did launch GPT-6.1 Terra on October 26, 2026, and the price claim checks out against OpenAI's own pricing page: $1 per million input tokens and $5 per million output, which is half the price of GPT-6.1 Sol and half or better than the older GPT-5.6 Terra on the standard tier. The "3x faster" figure is also a real OpenAI statement, but OpenAI said Terra finishes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol, which are its larger and more expensive models, not faster than the model it replaces. That matters, because it means the speed gain comes from using a smaller model rather than from a new generation being faster. OpenAI measured this on its own benchmark, no outside group has reproduced it, and independent analysis found the advantage shrinks a lot at the highest reasoning setting, where Terra takes about as long as the flagship. The post also cites a coding benchmark score of 50.3% without mentioning that Terra still ranks below OpenAI's own Sol model and below a rival model on that same test. Overall the post is mostly accurate and sourced to real OpenAI material, with the main problem being that a comparison against bigger sibling models was restated as a comparison against earlier models.
OpenAI [drifted from the evidence:] launched GPT-6.1 Terra [drifted from the evidence:] with 50% lower API prices and [drifted from the evidence:] 3x faster [drifted from the evidence:] computer use than [drifted from the evidence:] previous models.
OpenAI [added by the neutral restatement:] has released a model called GPT-6.1 Terra. [added by the neutral restatement:] Its API prices [added by the neutral restatement:] are 50% lower than some comparison model, and [added by the neutral restatement:] it performs computer-use tasks about three times faster than [added by the neutral restatement:] earlier models.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- OpenAI released GPT-6.1 Terra on October 26 2026. This is confirmed on OpenAI's blog, its X account, its help-center release notes and its API documentation.
- The model is available through the API as gpt-6.1-terra, and in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users.
- Standard API pricing is $1.00 per million input tokens, $0.05 cached input and $5.00 output, confirmed on OpenAI's own pricing page.
- Against GPT-6.1 Sol at $2 input and $10 output, that is exactly a 50% cut on both. Against GPT-5.6 Terra at $2 input and $12 output, input is 50% lower and output is about 58% lower, so the "50%" headline is accurate on input and understates the output cut.
- OpenAI did state that Terra finishes tasks over three times faster on average than GPT-6 Astra and GPT-6.1 Sol on its Agents' Last Exam task set. The post is reporting a real vendor statement, not inventing one.
- The supporting figures in the post's caption match OpenAI's published numbers: the 15-percentage-point gain over GPT-5.6 Terra at medium effort, the error rate falling from 53.6% to 16.7% at low effort and to 4.3% at max, and 50.3% on Terminal-Bench 4.0 against 21.5%.
- The headline says three times faster "than previous models," but OpenAI's comparators for that figure are GPT-6 Astra, its flagship, and GPT-6.1 Sol, the tier above Terra. OpenAI published no three-times speed figure against GPT-5.6 Terra, the model Terra actually replaces. The underlying fact is that a smaller, cheaper model runs faster than larger ones, which is a tier difference. The headline wording invites readers to hear it as a generation-over-generation speedup, which is a different and unsupported statement.
- "over 3x faster" is an average over OpenAI's own Agents' Last Exam tasks and is sensitive to the reasoning effort setting. One analysis of the launch reports that at maximum effort Terra takes roughly as long as Astra, so the speed advantage is not a fixed property of the model.
- Every performance figure in the post, including the speed multiple, the benchmark scores and the error rates, was produced by OpenAI. No independent runner has reproduced the Agents' Last Exam result. The post labels the source as OpenAI but presents the numbers in the voice of settled fact.
- The 50% price cut applies to the standard service tier. OpenAI's pricing page also lists a Fast tier for the same model at $2.00 input and $10.00 output, and requests above 272K input tokens bill at a higher rate for the entire request. A developer on those paths does not see a 50% reduction against GPT-5.6 Terra's standard rate.
- The caption reports Terminal-Bench 4.0 as 50.3% against the predecessor's 21.5% but omits that on the same comparison Terra sits below GPT-6.1 Sol and well below Claude Sonnet 5.5. One analysis notes Terra at maximum effort scores 50.3% on Terminal-Bench 4.0, below GPT-6.1 Sol's 55.9% and Claude Sonnet 5.5's 70.6%. OpenAI itself directs complex coding to the larger models.
- OpenAI's Terminal-Bench 4.0 figures and Artificial Analysis's figures for the same benchmark differ substantially, and AA's run uses a specific agent harness, task count, repeat count and step cap. The two sets of numbers should not be read as interchangeable.
- The caption's illustrative tasks, turning architectural drawings into 3D models and reviewing video, are descriptions of items inside OpenAI's benchmark set. They are evidence of benchmark performance under an evaluation harness, not of reliable shipping-product behaviour under ordinary user conditions.
- Whether the three-times speed advantage holds on workloads outside OpenAI's own task set. The independent figures reported from Artificial Analysis point to a smaller margin against GPT-6.1 Sol than three times on both throughput and seconds per task, though the workload is different and the two measurements are not directly comparable.
- Whether Terra is faster than GPT-5.6 Terra at all, and by how much. No published figure addresses that pairing.
- The standing of Agents' Last Exam as a benchmark: its set size, version, scoring rule and whether any party other than OpenAI can run it were not established by the evidence found.
- Whether OpenAI's "50% lower standard API prices" statement is meant as input-only or as a rounded description of the whole card. The published numbers give 50% on input, 75% on cached input and about 58% on output against GPT-5.6 Terra.
OpenAI did release GPT-6.1 Terra on October 26 2026. The model appears on OpenAI's own blog, its X account, its help-center release notes, and its API docs under the identifier gpt-6.1-terra. OpenAI's post states the model is rolling out and is available to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, with developer access through the API as gpt-6.1-terra. On price, OpenAI's pricing page lists gpt-6.1-terra at $1.00 input, $5.00 output, $0.05 cached input per million tokens at standard rates, with batch at $0.50 input and $2.50 output. The predecessor's docs page lists GPT-5.6 Terra at $2.00 input, $0.20 cached input and $12.00 output per million tokens. So input is exactly 50% lower, cached input is 75% lower, and output is about 58% lower. Against GPT-6.1 Sol at $2 input and $10 output, the cut is exactly 50% on both. On speed, OpenAI's own wording names its comparators. OpenAI posted that Terra "substantially outperforms GPT-5.6 Terra on Agents' Last Exam," and that "It even matches GPT-6 Astra at one-tenth the price per token and finishes tasks over 3× faster on average than both Astra and GPT-6.1 Sol." OpenAI's developer account framed the same figure as Terra delivering Astra-level performance on Agents' Last Exam while completing tasks over 3x faster, at one-tenth the cost per token. Independent measurement exists and is more modest than the headline. Artificial Analysis scores GPT-6.1 Terra at max effort 48 on its Intelligence Index against 52 for GPT-6.1 Sol and 53 for GPT-6 Astra, with GPT-5.6 Terra at 42. One analyst write-up relaying AA's figures notes that AA measured Terra at 91 output tokens per second at medium effort against Sol's 55, and 63 seconds per task in its index run against Sol's 149, adding that OpenAI's "more than 3x faster" claim covers OpenAI's own task set only, not every workload. Another notes that the "3x faster" claim depends on effort: at medium effort a task takes about a minute, while at max effort Terra takes about as long as Astra. Press coverage flagged the vendor-run provenance directly: The Next Web reported that Terra completed tasks at roughly a ninth of the cost, "though these results come from OpenAI's own testing and have not been independently checked."
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Did OpenAI actually release a model called GPT-6.1 Terra?
Yes. It was released on October 26, 2026, and is confirmed on OpenAI's blog, its X account, its help-center release notes, and its API documentation.
Is the claim of 50% lower API prices accurate?
Yes for input tokens. Standard pricing is $1.00 per million input tokens compared with $2.00 for GPT-6.1 Sol and GPT-5.6 Terra, an exact 50% cut, though output pricing is cut by about 58% against GPT-5.6 Terra, so the 50% figure slightly understates that part. This only applies to the standard tier; a separate Fast tier is priced higher.
Is Terra really three times faster than previous models?
OpenAI did say Terra finishes tasks over three times faster, but the comparison was to GPT-6 Astra and GPT-6.1 Sol, which are larger sibling models, not to GPT-5.6 Terra, the model it actually replaces. No published figure compares Terra's speed directly to GPT-5.6 Terra.
Has anyone outside OpenAI confirmed the speed and benchmark numbers?
No independent group has reproduced OpenAI's Agents' Last Exam result. Independent analysis from Artificial Analysis found a smaller speed advantage over GPT-6.1 Sol than claimed, and noted the gap shrinks a lot at the highest reasoning setting, where Terra takes about as long as Astra.
Is the coding benchmark score of 50.3% the full picture?
No. That score is accurate but the claim omits that Terra still ranks below OpenAI's own GPT-6.1 Sol model, which scored 55.9%, and well below Claude Sonnet 5.5, which scored 70.6%, on the same Terminal-Bench 4.0 test.