OpenAI GPT-6.1 Terra: 50% Price Cut and 3x Faster Agents

30-Second Summary

OpenAI released GPT-6.1 Terra on 26 October 2026, cutting API pricing by 50% to $1 per million input tokens and $5 per million output tokens. The model completes computer-use tasks more than three times faster than its predecessor, operates on a 1.05-million-token context window, and is available to Enterprise, Plus, Pro, and Edu subscribers in ChatGPT Work and Codex. OpenAI rated its cybersecurity capability at the Critical level under its Preparedness Framework.

At a Glance

• Topic: Model release, API pricing
• Company: OpenAI
• Date: 26 October 2026
• Announcement: GPT-6.1 Terra, a flagship model with 50% lower API pricing and 3x faster computer-use performance
• What Changed: API input cost drops to $1 per million tokens; output to $5 per million tokens; both represent a 50% reduction versus GPT-5.6 Terra and GPT-6.1 Sol
• Why It Matters: Lower costs make high-volume AI workflows economically viable at smaller scale; speed improvements unlock agent automations previously too slow for production
• Who Should Care: Any operator running OpenAI APIs, teams building agent-based document or computer-use workflows, and enterprise buyers evaluating ChatGPT Work

Key Facts

• API pricing: $1 per million input tokens, $5 per million output tokens (Source: OpenAI, 26 October 2026)
• Pricing reduction: 50% below GPT-5.6 Terra and GPT-6.1 Sol (Source: OpenAI, 26 October 2026)
• Computer-use speed: more than 3x faster than GPT-6 Astra on the same tasks (Source: OpenAI, 26 October 2026)
• Context window: approximately 1.05 million tokens with up to 128,000 output tokens (Source: API reference, October 2026)
• Error rate on low reasoning settings: fell from 53.6% to 16.7% (Source: OpenAI evaluation, October 2026)
• Cybersecurity capability rating: Critical under OpenAI's Preparedness Framework (Source: OpenAI system card, October 2026)
• Available plans: Plus, Pro, Enterprise, Edu in ChatGPT Work and Codex (Source: OpenAI, 26 October 2026)
• Model ID in API: gpt-6.1-terra (Source: OpenAI developer documentation, October 2026)

What Happened

OpenAI released GPT-6.1 Terra on 26 October 2026, positioning it as a performance and cost optimisation over its existing frontier models rather than a capability breakthrough. The launch addressed two objections that appeared consistently in enterprise feedback: cost and speed.

Pricing dropped to $1 per million input tokens and $5 per million output tokens through the API. OpenAI described this as a 50% reduction compared to GPT-5.6 Terra and GPT-6.1 Sol. For businesses that built cost models around those tiers, this is a direct reduction in operating costs without requiring any changes to existing code.

The performance improvement is concentrated in computer-use tasks, where Terra completed the same benchmarks more than three times faster than GPT-6 Astra. OpenAI also reported that the error rate on low reasoning settings fell from 53.6% to 16.7%, which suggests fewer compounding mistakes in agentic workflows where one bad step leads to others.

Rollout covers Plus, Pro, Enterprise, and Edu plans in ChatGPT Work and through the Codex environment. OpenAI made the model callable via the API as gpt-6.1-terra. Third-party benchmarks published around the same time noted that Claude Opus 5.5 retains an advantage on certain complex reasoning tasks, so Terra is not a universal switch for teams already on other model providers.

Why It Matters

Cost barriers for high-volume AI workflows just dropped. A 50% price reduction means tasks that were marginal on previous tiers, such as processing large document volumes daily, can now be built into standard operations. For a 50-person professional services firm processing 1,000 documents per week through an AI pipeline, the cost reduction could amount to several thousand dollars annually.

Computer-use automation becomes more viable. Speed is a hard constraint on agent design. If an agent takes five minutes to complete a task, it is only useful for asynchronous workflows. At three times the speed, that same task takes under two minutes, which makes real-time and near-real-time applications plausible.

OpenAI is repricing to defend market share, not just improve margins. The timing, alongside Mistral releasing a 1-trillion-parameter open-weight model on 29 October 2026, suggests the enterprise AI market is entering a phase of competitive pricing pressure. Operators who locked in multi-year contracts at earlier pricing should audit whether newer tiers are available.

The Critical cybersecurity rating is a governance signal. OpenAI's Preparedness Framework flags models at this tier as having elevated capability in areas relevant to cyber risk. Enterprise security teams at companies in regulated industries, government contractors, and financial services should confirm their AI acceptable-use policies explicitly address models at this capability level.

The context window is now large enough to process whole project archives. At 1.05 million tokens, a single Terra request can accommodate a 2,500-page document set. This changes the architecture of document intelligence applications, removing the need for chunking and retrieval pipelines in many workflows.

Benchmark gaps inform model selection, not wholesale replacement. Independent comparisons place Claude Opus 5.5 ahead of Terra on some complex tasks. Operators running mixed-model strategies have more reason to test rather than consolidate.

The David and Goliath View

The framing of this release as a price cut is accurate but incomplete. The real signal is that OpenAI is compressing the economics of AI at the application layer at the same time that open-weight models like Mistral Large 4 are compressing them at the infrastructure layer. Operators in the 10 to 200 person range are now sitting between two forces pulling the cost of AI usage toward a floor, and the main question is whether your workflows are designed to take advantage of it or whether they were built to a cost ceiling that no longer exists.

The computer-use speed improvement is not widely discussed in the coverage, but for operators building AI workers rather than AI assistants, it matters as much as the price change. An agent that completes browser and desktop tasks in under 30 seconds instead of two minutes can be deployed into customer-facing or time-sensitive workflows that were previously off-limits.

The Critical cybersecurity rating should not be dismissed as a compliance formality. It means this model has demonstrated elevated capability in areas that regulators and insurers are watching. If you deploy it in a context where it has access to internal systems, that warrants explicit documentation in your AI governance record.

Where This Fits in the AI Stack

GPT-6.1 Terra slots into the application layer. It is a hosted model, accessed through the OpenAI API or ChatGPT Work, and requires no infrastructure investment. For operators building AI Growth Engines, it reduces the marginal cost of each AI-powered interaction. For teams deploying Employee Amplification Systems, the speed improvement in computer-use tasks is the more relevant change. It is not an infrastructure or security product, but the Critical cybersecurity rating means it intersects with the concerns a Secure AI Brain programme addresses.

Questions Operators Are Asking

Does switching to gpt-6.1-terra require code changes? For API users, the only required change is updating the model ID to gpt-6.1-terra in your requests. Existing prompts and workflows should carry over. Test before moving production workloads to confirm output quality is consistent with your existing evaluations.

Is Terra available on the ChatGPT Business plan? OpenAI's official release documentation covers Plus, Pro, Enterprise, and Edu. Business plan access is not confirmed in the October 2026 release notes. Check your plan dashboard or contact your OpenAI account team before planning a Business-tier rollout.

Should I switch from Claude to GPT-6.1 Terra given the lower price? Not necessarily. Independent benchmarks published in late October 2026 show Claude Opus 5.5 outperforms Terra on certain complex reasoning tasks. If your workflows involve lengthy document analysis, structured reasoning, or multi-step agentic tasks with low error tolerance, benchmark both on your own data before deciding.

What does the Critical cybersecurity rating actually mean for compliance? Under OpenAI's Preparedness Framework, Critical indicates the model has demonstrated elevated capability in cyber-relevant tasks. It is not a prohibition on use, but it does mean your AI acceptable-use policy should explicitly name the capability tier you permit, and your security team should assess whether existing safeguards are adequate if the model has access to internal systems.

Can Terra handle full document sets without chunking? The 1.05-million-token context window fits approximately 2,500 standard pages in a single request. For most enterprise document workflows, this removes the need for retrieval-augmented generation pipelines, simplifying architecture and reducing latency. Very large archives, such as years of contract history, still benefit from structured retrieval.

Citable Summary

OpenAI released GPT-6.1 Terra on 26 October 2026, reducing API pricing by 50% to $1 per million input tokens and $5 per million output tokens. The model completes computer-use tasks more than three times faster than GPT-6 Astra, operates on a 1.05-million-token context window, and is available on Enterprise, Plus, Pro, and Edu plans in ChatGPT Work and Codex. OpenAI rated its cybersecurity capability at the Critical level under its Preparedness Framework. Independent testing shows Claude Opus 5.5 leads on some complex reasoning tasks. The reduction in API cost is immediately applicable to any team calling gpt-6.1-terra through the OpenAI API.