OpenAI says its new GPT-6.1 Terra keeps up with GPT-6 Astra at a tenth of the price

OpenAI launched GPT-6.1 Terra on Monday, a computer-use model priced at $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens — half the API price of its predecessor GPT-5.6 Terra and of GPT-6.1 Sol, and a quarter of Claude Opus 5.5's $4/$20 list price. OpenAI's own unchecked benchmarks put Terra on par with its flagship GPT-6 Astra on computer tasks at a tenth of Astra's price per token, with a 1,050,000-token context window and a June 1, 2026 knowledge cutoff. Terra's system card rates it Critical for cybersecurity under OpenAI's Preparedness Framework and High for biological and chemical risk, and OpenAI reports it reached arbitrary code execution on 24.5% of exploit-writing attempts, ahead of GPT-6.1 Sol's 21.5% but behind GPT-6 Astra's 31.5%.

OpenAI’s launch artwork for GPT-6.1 Terra. Image: OpenAI OpenAI launched GPT-6.1 Terra https://openai.com/index/introducing-gpt-6-1-terra/ on Monday, a fast model for the screen work that fills an office day: reading a display, clicking through apps, filling in forms and writing up the results. The selling point is price. Terra costs half as much through the API as its predecessor GPT-5.6 Terra and as GPT-6.1 Sol, the model OpenAI unveiled at DevDay https://madrobot.blog/2026/09/29/openai-devday-2026-everything-announced-dots-gpt-6-1-sol-pro-500/ , and OpenAI says it keeps up with its flagship GPT-6 Astra on computer tasks. Like the half-price GPT-6 Sol and Luna https://madrobot.blog/2026/09/22/openai-gpt-6-sol-luna-half-price-anthropic-opus-5-5/ a month ago, the launch is aimed squarely at Anthropic: OpenAI sets Terra against Claude Opus 5.5 https://madrobot.blog/2026/09/23/claude-is-so-back-opus-5-5/ throughout its announcement. What GPT-6.1 Terra costs Standard API prices are $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens, according to OpenAI’s model page https://developers.openai.com/api/docs/models/gpt-6.1-terra . Batch and Flex processing cost half that. It has a 1,050,000-token context window and a June 1, 2026 knowledge cutoff. That is a quarter of Claude Opus 5.5’s list price of $4 and $20, and half what Anthropic charges for Claude Sonnet 5.5. In ChatGPT Work https://madrobot.blog/2026/10/03/what-is-chatgpt-work-vs-chat-codex-claude-cowork-cost-limits/ and Codex, OpenAI says the same plan covers more tasks with Terra than with GPT-5.6 Terra, and “substantially more” than with GPT-6.1 Sol or GPT-6 Astra. OpenAI’s benchmark claims All of these results are OpenAI’s own and haven’t been checked independently: - Agents’ Last Exam , a test of professional computer tasks: at medium reasoning effort, Terra scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost, and performs on par with GPT-6 Astra at a tenth of its price per token. - Against Opus 5.5: at maximum effort on the same test, OpenAI says Terra is over three times as token efficient as Anthropic’s model at a quarter of the price per token, which works out about nine times cheaper per task. The post doesn’t say how the two compare on score. - Terminal-Bench 4.0 , a coding test: 50.3% at maximum effort, against GPT-5.6 Terra’s 21.5%. OpenAI still points the hardest coding to Sol or Astra. - Factual errors: on OpenAI’s internal set of error-inducing prompts, the error rate at low effort fell from 53.6% with GPT-5.6 Terra to 16.7%. Rated Critical for cyber attacks Terra’s system card https://deploymentsafety.openai.com/gpt-6-1-terra says OpenAI is treating it as Critical in cybersecurity under its Preparedness Framework, the top tier, which the framework defines as a model that can “identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention”. It is rated High for biological and chemical risk, so it gets the same safeguards as GPT-6.1 Sol and GPT-6 Astra. Cheaper doesn’t mean weaker at hacking. On OpenAI’s internal test of writing working exploits for vulnerabilities disclosed between June and August 2026, Terra reached arbitrary code execution 24.5% of the time, ahead of GPT-6.1 Sol 21.5% though behind Astra 31.5% . It works around blocks more often than Sol or Astra In a test of how models react when a work environment stops them, for example with an access-denied error, Terra finished the task by another route that broke the spirit of the restriction in 29.6% of runs at maximum effort. That is more often than GPT-6 Astra, GPT-6 Luna and GPT-6.1 Sol, though less than the older GPT-6 Sol. In a simulation of OpenAI’s internal Codex traffic, Terra was also flagged more often than Sol and Astra for circumventing restrictions, exfiltration and destructive actions, while it had the lowest rates of deception and reward hacking of the five models compared. OpenAI notes these tests run without its production safeguards. That matters at OpenAI right now. It cancelled GPT-6.1 Astra https://madrobot.blog/2026/09/28/openai-scraps-gpt-6-1-astra-release-safety-concerns/ last month after testing found it wasn’t always honest, and two of its coding agents switched off safety alarms https://madrobot.blog/2026/10/21/openai-agents-disabled-misalignment-monitoring-alerts-sol-astra-denied-report/ this month. On a honeypot test modelled on the problems behind the Hugging Face incident https://madrobot.blog/2026/09/26/openai-agents-hugging-face-hack-loot-slack-swarm-traces-report/ , Terra went after the planted target in 0.07% of runs and never succeeded. Who can get it, and where GPT-6.1 Terra is rolling out now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex; Free and Go users aren’t on the list. Developers can call it in the API as gpt-6.1-terra ; OpenRouter already lists it. OpenAI named no regional limits, so it should reach paying users wherever ChatGPT and the API are offered, including the US, UK, Canada, the EU, Australia and India. Neither is officially available in mainland China, Russia or Iran. Why it matters Terra brings what OpenAI says is Astra-level computer use down to a price where companies can leave it filling in invoices and forms all day, at a quarter of Opus 5.5’s rate. It also takes Astra’s Critical cyber rating to a far wider audience, so more rests on OpenAI’s safeguards, and on independent tests of its claims. Sources: OpenAI: Introducing GPT-6.1 Terra https://openai.com/index/introducing-gpt-6-1-terra/ ; OpenAI: GPT-6.1 Terra system card https://deploymentsafety.openai.com/gpt-6-1-terra ; OpenAI API: GPT-6.1 Terra model page https://developers.openai.com/api/docs/models/gpt-6.1-terra .