OpenAI launches GPT-6.1 Terra, a cheaper model for computer use and agentic work

Overview

Updated Oct 29

OpenAI released GPT-6.1 Terra on October 26, a model built for the screen-based work that fills office days: reading invoices, filling forms, editing videos and interpreting technical drawings. Standard API pricing is $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% below GPT-5.6 Terra.

The bet is that cost, more than raw capability, will decide how fast AI agents move into routine back-office work. OpenAI says Terra performs some professional tasks at a level comparable to its flagship GPT-6 Astra, at one-tenth the token price. Its benchmarks also show it completing tasks more than three times faster than the larger models. These figures come from OpenAI's own tests and have not been independently verified.

Why it matters

If Terra's price-performance holds in production, automating high-volume office work becomes cheap enough to change how businesses staff routine tasks.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

API input price, per million tokens

Standard input rate for GPT-6.1 Terra.

API output price, per million tokens

Standard output rate for GPT-6.1 Terra.

50%

List-price cut vs. GPT-5.6 Terra

OpenAI's stated reduction in standard API pricing.

15 pts

Agents' Last Exam gain at medium reasoning

Terra's score over GPT-5.6 Terra at roughly one-third the cost, per OpenAI.

16.7%

Low-effort error rate on OpenAI's internal benchmark

Down from 53.6% for GPT-5.6 Terra on error-inducing prompts.

Claimed speedup over GPT-6 Astra and GPT-6.1 Sol

OpenAI's estimate of how much faster Terra completes computer-use tasks.

Voices

Curated perspectives — historical figures and your fellow readers.

People Involved

Sam Altman

Chief Executive Officer, OpenAI

Leading OpenAI through the GPT-6 model family rollout

Organizations Involved

OpenAI

Artificial intelligence company

Developer of GPT-6.1 Terra

Released GPT-6.1 Terra on October 26, 2026 as its lower-cost computer-use model.

Amazon Web Services (AWS)

Cloud computing provider

Availability partner for GPT-6.1 Terra on Bedrock

Made GPT-6.1 Terra generally available on Bedrock the same day as OpenAI's launch.

Microsoft

Technology company

Availability partner for GPT-6.1 Terra in Foundry and Azure AI

Made GPT-6.1 Terra generally available in Foundry and Azure AI for agents that see and act.

GitHub

Software platform, subsidiary of Microsoft

Availability partner for GPT-6.1 Terra in Copilot

Added GPT-6.1 Terra to Copilot for everyday agentic coding and terminal work.

Timeline

2026 October 2026

4 events Latest: October 26th, 2026 · 2 weeks ago

Tap a bar to jump to that date

October 2026
OpenAI launches GPT-6.1 Terra
Latest Product Release

Lower-cost computer-use model launches at $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% below GPT-5.6 Terra.
October 26th, 2026
AWS adds Terra to Bedrock
Availability

Enterprise customers can call the model through Amazon's managed API platform.
October 26th, 2026
Microsoft and GitHub roll out Terra
Availability

Foundry, Azure AI, and GitHub Copilot all gain access to the model the same day.
October 26th, 2026
2026
OpenAI ships GPT-6 Astra and GPT-6.1 Sol
Product Release

The flagship frontier model and the coding-focused model anchor the GPT-6 family.
2026

Scenarios

GPT-6.1 Terra becomes OpenAI's most-used model

Likely Resolves by Q2 2027

Discussed by: OpenAI, framing Terra as its workhorse for high-volume automation

If Terra's price and speed hold up in production, it could displace GPT-5.6 Terra and take a large share of agentic API traffic. The cost-per-task advantage OpenAI claims is the core driver. Adoption would show up in token volumes that OpenAI occasionally reports in developer updates.

Anthropic or Google matches Terra's price-performance

Possible Resolves by Q1 2027

Discussed by: Competitive analysts tracking the agent-pricing race

Terra sets a new price floor for computer-use models. A rival like Anthropic or Google could respond by shipping a comparable model at or below $1 per million input tokens. A matching launch would signal that Terra's pricing, not its capability, is the harder advantage to hold.

A documented exploit emerges in Terra's computer-use mode

Possible Resolves by Q3 2027

Discussed by: Security researchers; OpenAI's own prompt-injection testing

Computer-use models that act on screens are exposed to malicious instructions hidden in web pages and documents. OpenAI says Terra showed strong resistance in its own prompt-injection tests, but real-world deployment is a harder proving ground. A demonstrated takeover would prompt safeguards updates or feature restrictions.

Historical Context

2 moments from history that rhyme with this story — and how they unfolded.

March 2023

GPT-3.5 Turbo (2023)

OpenAI released a lightweight, much cheaper version of its GPT-3.5 model as an API. Priced far below the flagship, it became the default engine for thousands of chatbot startups within months.

Then

The cost collapse triggered a wave of consumer and developer AI products built on the API.

Now

It established that price, not raw capability, drove the mass adoption of AI APIs.

Why this matters now

Terra is the same lever being pulled again: a dramatic price cut on a capable model, aimed at making agentic workloads economical at scale.

March 2024

Claude 3 Haiku (2024)

Anthropic shipped the smallest and cheapest model in its Claude 3 family, aimed at high-volume, low-latency tasks where a flagship model was overkill.

Then

Haiku gave Anthropic a competitive entry point for cost-sensitive automation customers.

Now

It popularized the three-tier family structure, fast and cheap, balanced, and flagship, that OpenAI's GPT-6 lineup now mirrors.

Why this matters now

Terra occupies the same tier for OpenAI, competing directly for budget-conscious automation buyers against Anthropic's small models.

Sources

(12)