Editorial illustration for GPT-6.1 Terra Halves OpenAI API Prices, Gains on Larger Models
OpenAI Halves GPT-6.1 Terra API Prices
OpenAI cut API prices for its mid-range model in half and claims the new version still beats its predecessor on performance. GPT-6.1 Terra, released this week, charges $1 per million input tokens and $5 per million output tokens, down from $2 and $10 for GPT-5.6 Terra. Cached inputs now run $0.05 per million tokens, a quarter of the old rate.
That pricing puts Terra directly against Anthropic's Claude Sonnet 5.5, which charges $2 for input and $12 for output, at less than half the cost. OpenAI built the model for office work and automated software control: drafting proposals, reading technical drawings, filling out forms. It pairs visual reasoning with what the company describes as more precise computer control and a less robotic writing style.
The model is live now for Plus, Pro, Enterprise, and Edu subscribers inside ChatGPT Work and Codex, plus the standard API. OpenAI frames the release as a shift in the price-to-performance curve rather than a straight upgrade, betting that users care more about cost per task than raw benchmark scores. Early third-party testing backs up at least part of that claim.
Artificial Analysis calls GPT-6.1 Terra a new benchmark for intelligence per dollar. At maximum reasoning effort, the model gains 6 points on the Intelligence Index over GPT-5.6 Terra while costing less than a third per task ($0.40 versus $1.40). At lower effort levels the gains are even larger—7 to 13 points—placing the model on the Pareto frontier, the point where cost and performance can no longer both improve simultaneously.
Why this matters
For teams building on OpenAI's API, Terra's pricing shift changes the math on what used to require a larger model. If a mid-range model can nearly match GPT-6 Astra and GPT-6.1 Sol on MMMU-Pro while beating GPT-5.6 Terra's best OSWorld 2.1 result at a fraction of the cost, the calculus for agent-heavy products, think desktop automation, office workflows, customer-facing tools, tilts toward building on the cheaper tier by default rather than reaching for the flagship. That matters most for founders running thin margins on API-dependent products and for researchers benchmarking cost-per-task rather than raw capability.
We'd still want independent verification of the hallucination claims before treating them as settled; OpenAI's own benchmarks are the source here, not a third party. Worth watching: whether rivals like Anthropic or Google respond with their own price cuts on comparable mid-tier models, and whether "automated software control" use cases hold up outside curated benchmark conditions once developers start running Terra against messier, real-world desktop tasks at scale.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google to End Free Access to Gemini AI for Non-Subscribers
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source
Related Topic OpenAI Developer: AI Agent "Swarms" Waste Tokens, Create "Coordination Tax
AI Tools & Apps
From Archives Nace AI's Drex posts 92.05% on BrowseComp-Plus with 12.6 search calls
From Archives Tsinghua AI Open-Sources AKV Cache Server for LLM Agents on SGLang
From Archives OpenAI reports promising results migrating COBOL code to modern languages
Policy & Regulation
From Archives Eric Schmidt compares OpenAI and Anthropic to colonial trading companies
Business & Startups
Common Questions Answered
How much did OpenAI reduce prices for GPT-6.1 Terra compared to GPT-5.6 Terra?
OpenAI cut API prices for GPT-6.1 Terra in half, reducing input token costs from $2 to $1 per million tokens and output token costs from $10 to $5 per million tokens. Additionally, cached inputs now cost $0.05 per million tokens, which is a quarter of the previous rate for GPT-5.6 Terra.
How does GPT-6.1 Terra's pricing compare to Anthropic's Claude Sonnet 5.5?
GPT-6.1 Terra is significantly cheaper than Claude Sonnet 5.5, costing less than half the price. Claude Sonnet 5.5 charges $2 for input and $12 for output tokens, while GPT-6.1 Terra charges $1 and $5 respectively, making it a more cost-effective option for API users.
What performance gains does GPT-6.1 Terra achieve over its predecessor according to Artificial Analysis?
At maximum reasoning effort, GPT-6.1 Terra gains 6 points on the Intelligence Index over GPT-5.6 Terra while costing less than a third per task ($0.40 versus $1.40). At lower effort levels, the gains are even larger at 7 to 13 points, placing the model on the Pareto frontier where cost and performance can no longer both improve simultaneously.
Why does GPT-6.1 Terra's pricing change the development strategy for agent-heavy products?
With GPT-6.1 Terra nearly matching larger models like GPT-6 Astra and GPT-6.1 Sol on benchmarks while costing significantly less, developers building agent-heavy products such as desktop automation, office workflows, and customer-facing tools can now default to the cheaper mid-range tier instead of reaching for flagship models. This pricing shift fundamentally changes the cost-benefit calculation for teams building on OpenAI's API.
What specific benchmark improvements does GPT-6.1 Terra demonstrate?
GPT-6.1 Terra achieves competitive performance on MMMU-Pro benchmarks compared to larger models and beats GPT-5.6 Terra's best OSWorld 2.1 result at a fraction of the cost. These improvements demonstrate that the mid-range model can handle complex reasoning tasks that previously required more expensive flagship models.
Further Reading
- Introducing GPT-6.1 Terra - OpenAI
- Addendum to GPT-6 Astra System Card: GPT-6.1 Terra - OpenAI Deployment Safety Hub
- GPT-6.1 Terra: Benchmarks, Pricing & Model Comparisons - Kingy AI
- GPT-6.1 Terra (Non-reasoning) - Intelligence, Performance & Price Analysis - Artificial Analysis
- GPT-6.1 Terra launches with 50% lower standard API prices than GPT-5.6 Terra - Digg
