A new Terra model built for everyday agent work
OpenAI has released GPT-6.1 Terra, positioning it as a fast, efficient model for visual reasoning, precise computer use, coding assistance, and day-to-day professional writing . The launch matches the BleepingComputer story: this is not a flagship-only model announcement, but a cost-and-speed upgrade aimed at the workhorse tier of OpenAI’s lineup, where developers and businesses care less about peak benchmark scores and more about how many useful tasks an agent can finish per dollar .
The company says GPT-6.1 Terra is now available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and developers can access it through the API as gpt-6.1-terra . OpenAI’s developer community announcement also framed the release around computer use: the model can understand what is on screen, take action across apps, and write more naturally for routine professional work . That framing matters because the “agent” market is moving away from simple chat responses toward systems that can operate browsers, terminals, files, forms, and business applications.
The headline numbers: 3x faster and 50% cheaper
The most important commercial claim is pricing. OpenAI says GPT-6.1 Terra’s standard API prices are 50% lower than those of GPT-5.6 Terra and GPT-6.1 Sol . The listed standard rates are $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens . BleepingComputer reported the same pricing cut and emphasized that it makes Terra significantly less expensive than larger OpenAI models for repetitive work .
The second headline claim is speed. On OpenAI’s Agents’ Last Exam benchmark, Terra reportedly completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol . AWS repeated that comparison in its own availability announcement for Amazon Bedrock, saying GPT-6.1 Terra completed tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol in OpenAI’s evaluation . The key caveat is that these numbers are based on OpenAI’s benchmark environment, not a broad set of independent production tests.
Why this matters for autonomous agents
A 50% list-price cut changes the economics of agentic software because agents often burn tokens while planning, observing, retrying, and verifying their work. A chatbot may answer once; an agent may read a screen, decide what to click, inspect the result, call tools, correct mistakes, and produce a final report. In that pattern, even small pricing differences can become large monthly cost differences.
OpenAI’s examples show the target market clearly: invoice processing, filling out forms, interpreting technical drawings, drafting proposals, reviewing media, and working through desktop applications . These are not just “ask-and-answer” tasks. They require a model to combine visual interpretation, instruction following, and action sequencing. If Terra can deliver acceptable reliability at lower cost, it could let companies automate more medium-complexity workflows without reserving premium models for every step.
This is also why the model’s position in the lineup is important. OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra for the most complex coding and research work . Terra is being sold as the practical middle: strong enough for many professional workflows, cheaper than Sol, and faster than larger models on the agent benchmarks OpenAI highlights.
Benchmark gains: agents, visual reasoning, and coding
OpenAI says GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra at medium reasoning effort on Agents’ Last Exam while costing about one-third as much . The company also says it performs on par with GPT-6 Astra at one-tenth the price per token on that evaluation . BleepingComputer highlighted the same result as evidence that Terra narrows the gap between OpenAI’s middle tier and its higher-end models .
For visual reasoning, OpenAI says Terra outperforms GPT-5.6 Terra at every reasoning setting on MMMU-Pro and nearly matches GPT-6 Astra and GPT-6.1 Sol at much lower cost . The model also surpasses GPT-5.6 Terra’s best score at medium reasoning effort on OSWorld 2.1, a benchmark designed to test how well agents use desktop apps for multistep tasks .
Coding is improved, but not presented as Terra’s strongest territory. OpenAI reports that GPT-6.1 Terra solves 50.3% of Terminal-Bench 4.0 tasks at maximum reasoning effort, more than double GPT-5.6 Terra’s published 21.5% score . BleepingComputer also reported those two figures and noted that OpenAI still points users to Sol or Astra for the most complex coding tasks .
Fewer factual errors, according to OpenAI
OpenAI is also pitching GPT-6.1 Terra as more reliable than its predecessor. On the company’s internal benchmark of error-inducing prompts, the error rate at low reasoning effort reportedly fell from 53.6% with GPT-5.6 Terra to 16.7% with GPT-6.1 Terra . At maximum effort, OpenAI says the error rate fell to 4.3%, nearly matching GPT-6 Astra and GPT-6.1 Sol . BleepingComputer described this as a notable hallucination reduction, while also relying on OpenAI’s own evaluation numbers .
That distinction matters. Internal evaluations are useful for understanding how OpenAI is positioning the model, but buyers will still need to test Terra in their own applications. The hardest production failures often appear in messy enterprise environments: malformed invoices, unusual file formats, brittle internal tools, ambiguous permissions, and real-world edge cases that no public benchmark fully captures.
Safety, prompt injection, and cybersecurity status
OpenAI published a system card addendum for GPT-6.1 Terra on October 26, 2026 . The safety document says the model is being treated as Critical in cybersecurity under OpenAI’s Preparedness Framework and High for biological and chemical capability . It also says Terra uses the same safeguards stack as GPT-6.1 Sol and GPT-6 Astra, with probes updated for compatibility .
The system card reports that GPT-6.1 Terra performs similarly to recent GPT-6 and GPT-6.1 releases on agentic safety evaluations . It also discusses prompt-injection testing, a key issue for computer-use agents because malicious instructions can be hidden in web pages, documents, emails, or application content the agent reads while completing a task . OpenAI says Terra is substantially robust to indirect prompt injections, but the company also notes that it is continuing to update these evaluations .
For businesses, the practical takeaway is that cheaper and faster agents do not remove the need for permissions, logging, approvals, sandboxing, and human review. A model that can act across apps is useful because it can do more; it is risky for the same reason.
Distribution expands through Amazon Bedrock
The launch is not limited to OpenAI’s own channels. AWS announced that GPT-6.1 Terra is generally available on Amazon Bedrock as of October 26, 2026 . AWS said customers can call the model directly through Bedrock APIs or configure ChatGPT Work and Codex to use the model on Amazon Bedrock . The company highlighted enterprise controls for securing workloads, governing access, and auditing model invocation activity .
That distribution could matter for adoption. Many enterprises already use AWS identity, governance, and monitoring tools, and Bedrock availability can make procurement and deployment easier than connecting directly to a separate model provider. AWS also said Terra supports a context window of up to 1 million input tokens and accepts both text and image inputs .
The market impact: lower-cost agents get more plausible
GPT-6.1 Terra is best understood as a pressure move on the economics of autonomous agents. It does not eliminate the need for top-tier models, and OpenAI is explicit that Sol and Astra remain better choices for the hardest coding and research tasks . But by lowering the cost of a model designed for computer use, OpenAI is making it more realistic for developers to run agents frequently rather than sparingly.
The current state of the story, based on sources published within the 72-hour window ending October 27, 2026, is clear: OpenAI has launched Terra, priced it aggressively, released safety documentation, and gained immediate cloud distribution through AWS . The next question is whether independent developers and enterprise pilots can reproduce the promised speed, reliability, and cost-per-task gains outside OpenAI’s own benchmarks.