OpenAI Releases GPT-6.1 Terra With Advanced Computer Use at One-Fourth the Cost of Claude

OpenAI launched GPT-6.1 Terra today, a mid-tier model built for precise computer use that it prices at $1 per million input tokens — a fraction of what competing flagship models charge. The release draws a direct line between cost and capability: OpenAI claims Terra performs 9x more tasks per dollar than Claude Opus 5.5, a comparison that will pressure enterprise buyers currently evaluating which AI to embed in their workflows.

What GPT-6.1 Terra Is

GPT-6.1 Terra is not a reasoning giant or a general-purpose upgrade. It is a purpose-built model for computer use — AI that can see a screen, interpret visual content, and operate software the way a human assistant would. It combines three capabilities:

• Detailed visual reasoning — reading and interpreting what is on screen
• Precise computer use — executing mouse clicks, keyboard inputs, and multi-step navigation across applications
• Natural writing — generating coherent, professional-quality text for documents and communications

The model sits in the "Terra" tier of OpenAI's naming scheme, below flagship reasoning models and positioned as the practical, cost-effective choice for businesses running high-volume automated tasks.

The Numbers

According to OpenAI, GPT-6.1 Terra:

• Solved 50.3% of Terminal-Bench 4.0 tasks, a standardized benchmark for computer-use performance. Its predecessor, GPT-5.6 Terra, solved 21.5% — a 134% improvement.
• Costs $1 per million input tokens, compared to significantly higher rates for Claude Opus 5.5.
• Claims a 9x cost-per-task advantage over Claude Opus 5.5 at equivalent task completion volumes.

Terminal-Bench 4.0 tests whether a model can execute real computing tasks autonomously — file management, application navigation, multi-step workflows — rather than just answer questions about them. A jump from 21.5% to 50.3% is meaningful, though it still means the model fails on roughly half of tested tasks.

Who Gets It and When

GPT-6.1 Terra is available now through two products:

• ChatGPT Work — for individual professionals and teams
• Codex — OpenAI's software engineering platform

Access is limited to Plus, Pro, Enterprise, and Edu subscribers. Free-tier users are not included at launch.

What This Means for Enterprise Buyers

The cost argument is the headline. Enterprise teams running thousands of automated computer-use tasks daily — whether in finance, legal, healthcare administration, or software development — face real unit economics when choosing an AI backbone. If GPT-6.1 Terra delivers comparable results at one-ninth the cost, that math is hard to ignore.

The competitive implication is direct: OpenAI is explicitly benchmarking against Anthropic's Claude Opus 5.5, a premium model positioned at the high end of the market. Terra does not claim to beat Opus 5.5 on every dimension — it claims to beat it on cost-per-task, which is the metric that matters most for high-volume deployments.

For software teams already using Codex, Terra slots in as the default computer-use engine, potentially replacing manual QA steps or test execution workflows.

The important caveat: a 50.3% success rate on Terminal-Bench 4.0 means Terra still fails on roughly half of benchmark tasks. For workflows requiring near-perfect reliability, human oversight remains essential.

What to Watch

OpenAI has not confirmed when free-tier access will open or whether Terra will roll into the API for third-party developers building their own computer-use applications. Anthropic and Google are likely to respond with their own cost-efficiency benchmarks. Independent replication of the Terminal-Bench 4.0 results — not from OpenAI's own testing — will be the real signal of whether the 50.3% figure holds under varied conditions.