GPT-6.1 Terra is here: OpenAI's everyday computer-use model lands on Amazon Bedrock at $1/$5 per million tokens

• Who: OpenAI, with distribution through Amazon Web Services on Amazon Bedrock.
• What: GPT-6.1 Terra, a mid-tier model for everyday computer work, succeeding GPT-5.6 Terra in the Terra line.
• When: Launched October 26, 2026, with general availability on Amazon Bedrock announced the same day.
• Price: $1 per million input tokens, $5 per million output tokens, and $0.05 per million cached input tokens; AWS says input and output prices are at least 50% lower than GPT-6.1 Sol and GPT-5.6 Terra.
• Context: Up to 1 million input tokens, with text and image inputs.
• Benchmarks (OpenAI-reported): 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam, on par with GPT-6 Astra, and more than 3 times faster than GPT-6 Astra and GPT-6.1 Sol; 50.3% on Terminal-Bench 4.0 at max effort versus 21.5% reported for GPT-5.6 Terra.

OpenAI has launched GPT-6.1 Terra, its new model for the everyday workday: software that understands what is on screen, takes action across apps, and writes more naturally than its predecessor. On the same day, AWS announced general availability of the model on Amazon Bedrock, putting OpenAI's mid-tier model inside the cloud console where enterprise teams already govern their workloads. It is not the flagship — that remains GPT-6 Astra — but it is the model OpenAI wants most teams to actually run.

Terra sits in the middle of OpenAI's lineup by design. It succeeds GPT-5.6 Terra, and OpenAI positions it for the high-volume work that does not need a frontier model: drafting, triage, and agents that operate across business systems. For in-depth research or complex coding, OpenAI itself points users to GPT-6.1 Sol or GPT-6 Astra. The pitch is cost-intelligence, not raw capability: a model good enough for the bulk of tasks, cheap enough that running agents all day stops being a budget conversation.

The price and the performance claims

The headline number is $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million. AWS says those input and output prices are at least 50% lower than GPT-6.1 Sol and GPT-5.6 Terra. In ChatGPT Work and Codex, OpenAI says a user's usage allowance goes further with GPT-6.1 Terra than with GPT-5.6 Terra, and far more than with GPT-6.1 Sol or GPT-6 Astra.

On performance, OpenAI reports that GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra on Agents' Last Exam, a test of how well agents use software at work, performed on par with GPT-6 Astra, and completed tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. AI Weekly's summary of the launch adds a second OpenAI-reported figure: 50.3% of tasks solved on Terminal-Bench 4.0 at max effort, versus 21.5% reported for GPT-5.6 Terra. Treat all of these as vendor claims for now: the announcements report OpenAI's own evaluations, not independent benchmarks, and the Bedrock announcement gives no evaluation setup or breakdown.

Where you can actually use it

GPT-6.1 Terra is rolling out to all Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and developers can call it through the OpenAI API as gpt-6.1-terra. On Bedrock, teams can call it directly through Amazon Bedrock APIs or configure ChatGPT Work and Codex to use the model on Amazon Bedrock. The model accepts text and image inputs with a context window of up to 1 million input tokens.

AWS's use cases are deliberately ordinary: drafting proposals, interpreting technical drawings, investigating codebases, and building agents that work across business systems. The Bedrock pitch is less about the model and more about the wrapper — established AWS controls to secure workloads, govern access, and audit model invocation activity. For enterprises already standardized on AWS, that is the point: the same model, inside their existing identity, networking, and audit controls.

Why the cheap tier matters

For most of the last two years, the interesting release was the biggest model. Terra inverts that logic: cost per task, not peak capability, is what decides whether an agent gets deployed at scale. A model that matches the flagship on an agent benchmark at a fraction of the token price — OpenAI puts it at one-tenth of Astra's price per token — changes which workflows are worth automating, even if the benchmark gap to Astra reappears on harder tasks.

The open questions are the usual ones for a vendor launch. Independent tests have yet to confirm the speed and benchmark claims, and OpenAI's naming — Terra, Sol, Luna, Astra, with version numbers that do not move in lockstep — will confuse buyers until the lineup settles. What is not in doubt is the direction: the frontier labs are now competing as hard on price per task as on capability, and Bedrock availability means that competition has arrived inside the enterprise cloud. The next data point to watch is not another launch post, but the first independent benchmark run.

Sources: OpenAI Developer Community announcement, AWS announcement (via GIXtools), AI Weekly.