OpenAI has released GPT-6.1 Terra. It is a fast mid-range model that OpenAI says combines detailed visual reasoning, precise computer use, and clearer writing. The big number is the price. Standard API prices are 50 percent lower than GPT-5.6 Terra and GPT-6.1 Sol. For a team running lots of inference, this changes the math.
In one line: OpenAI cut the price of its Terra line roughly in half and added stronger visual reasoning and computer use, aimed at repetitive office work and coding agents.
Pricing and positioning
OpenAI lists standard API pricing at $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. The Decoder compares that with $2 per million input and $12 per million output for GPT-5.6 Terra. Cached input tokens cost a quarter of the previous rate.
Simon Willison notes that Terra sits in a similar position to Anthropic's Sonnet models. It is not the cheapest or the most expensive option. OpenAI's Ted Sanders wrote in a Hacker News comment that Terra is probably better than Astra and Sol at technical communication, similar on vision and computer use but much faster, and a bit worse on most other things. For API developers it is a midrange option between Sol and Luna.
Benchmarks: computer use, vision, coding
OpenAI reports large gains on Agents' Last Exam, a benchmark of professional computer use tasks. At medium reasoning effort, GPT-6.1 Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost. It performs on par with GPT-6 Astra at one-tenth the price per token. At maximum effort, it is over three times as token efficient as Claude Opus 5.5 at one-fourth the price per token, which OpenAI says results in about nine times lower cost per task. It also completes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol.
On MMMU-Pro, a visual reasoning benchmark, Terra outperforms GPT-5.6 Terra at every reasoning setting and nearly matches GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1 for desktop agent tasks, it surpasses GPT-5.6 Terra's best score at medium effort for a fraction of the cost.
For coding, Terminal-Bench 4.0 at maximum reasoning effort shows GPT-6.1 Terra solving 50.3 percent of tasks, up from 21.5 percent for GPT-5.6 Terra. OpenAI still says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks. On Terminal-Bench-Science 0.1, Terra surpasses Sol at lower cost and comes within five points of Claude Opus 5.5 at less than one-fifth the estimated cost.
Hallucinations and safety
OpenAI says the error rate on its internal error-inducing prompt benchmark fell from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra at low reasoning effort. At maximum effort the error rate is 4.3 percent. Artificial Analysis reports Terra guesses only 38 percent of the time on the AA-Omniscience benchmark when it lacks an answer, down from 88 percent for the predecessor.
On safety, Terra meets the Critical threshold in cybersecurity under OpenAI's Preparedness Framework. It inherits protections from GPT-6 Astra. The Decoder reports that indirect prompt injection defense is 98.8 percent, up from 96.2 percent for GPT-5.6 Terra.
What I would watch
For my work at Latent, this model is most interesting for document-heavy automation. Reading invoices and entering details into accounting software is exactly the kind of task where a 50 percent price cut matters. A client who wants to process hundreds of PDFs might now get acceptable accuracy without paying frontier prices.
I would run a small pilot on invoice extraction and form filling with GPT-6.1 Terra and compare it against our current extraction stack. For coding agents, the Terminal-Bench jump is promising but OpenAI still points to Sol and Astra for complex work. So I would keep Terra for simpler command-line fixes and not switch the main agent yet. The speed improvement on computer use could help with repeated UI tasks, but I want to see it handle Indian invoice formats before recommending it to a client. Nothing urgent, just a controlled test.