GPT-6.1 Terra Is Live on Felo API: Limited-Time 70% Off
GPT-6.1 Terra is live on Felo API with a limited-time 70% discount. See the promotional price and how it compares with GPT-5.6 Terra.
OpenAI has just released GPT-6.1 Terra, and you can now use it through Felo API. For a limited time, Felo lists a 70% discount on its openai-mix route: $0.30 per million input tokens and $1.50 per million output tokens. Terra also improves on its predecessor at several demanding tasks.
What does that mean if you write for work, analyze images, build software, or pay for an API? Here are the capabilities, published test results, and costs that matter.
What is GPT-6.1 Terra?
GPT-6.1 Terra is OpenAI's efficient model for visual reasoning, computer use, and everyday writing. It accepts text and images and returns text. OpenAI's suggested jobs include matching product photos to catalog entries, reading technical drawings, acting on application screens, and drafting or revising documents. The model can also work with tools through the Responses API, but computer use requires an application or agent that actually gives it access to those tools. A plain text prompt does not make it operate your computer. OpenAI model documentation
The model has a 1,050,000-token context window and a 128,000-token maximum output, the same published limits as GPT-5.6 Terra. Its default reasoning effort is medium, with none, low, high, xhigh, and max also available. More effort may improve a difficult answer, but it can also increase latency and the number of billed output tokens. GPT-6.1 Terra specifications · GPT-5.6 Terra specifications
For ordinary users, the practical description is simple: Terra is a strong choice when the task has clear inputs and a checkable result. Give it the source document, image, screen, or code change you want it to work on, and say what a successful answer should contain. For open-ended research or a complex software project, OpenAI still recommends GPT-6.1 Sol or GPT-6 Astra. OpenAI model selection guide
GPT-6.1 Terra vs GPT-5.6 Terra: what changed?
OpenAI's launch data shows improvements in computer use, visual understanding, factual reliability, and coding. The figures below come from OpenAI's published evaluations, so treat them as useful comparisons under test conditions, not as a guarantee for every prompt. OpenAI launch announcement
| Measure | GPT-5.6 Terra | GPT-6.1 Terra | What it tells you |
|---|---|---|---|
| Standard input price, per 1M tokens | $2.00 | $1.00 | 50% lower input price |
| Standard output price, per 1M tokens | $12.00 | $5.00 | About 58% lower output price |
| Cached input price, per 1M tokens | $0.20 | $0.05 | 75% lower cache-read price |
| Context window | 1.05M tokens | 1.05M tokens | No increase in published context size |
| Factual error rate on difficult prompts, low effort | 53.6% | 16.7% | Fewer errors on OpenAI's internal test |
| Terminal-Bench 4.0, maximum effort | 21.5% | 50.3% | More than twice as many coding tasks solved |
Pricing is the published standard rate for requests within the normal pricing band. The factuality figures use OpenAI's internal error-inducing prompt set; the coding figures use Terminal-Bench 4.0. Sources: OpenAI launch, new model card, previous model card.
Better with screens and images
On Agents' Last Exam, a test of professional computer-based work, GPT-6.1 Terra at medium reasoning effort scored 15 percentage points above GPT-5.6 Terra at about one-third of the estimated cost per task. OpenAI also reports that the new Terra beats its predecessor at every tested reasoning setting on MMMU-Pro, a visual reasoning benchmark. On OSWorld 2.1, which measures multistep desktop tasks, its medium-effort result exceeded GPT-5.6 Terra's best reported score. OpenAI launch announcement
That matters when the source is a screenshot, a product photo, a form, or a drawing. A model that reads the visual material more accurately can make fewer wrong turns before an agent takes action. You should still review any action that changes records or files, especially when the source image is unclear.
Fewer factual slips, stronger scoped coding
OpenAI's internal test deliberately uses prompts likely to trigger mistakes. At low reasoning effort, GPT-6.1 Terra's error rate fell from 53.6% to 16.7%. At maximum effort, OpenAI reports a 4.3% error rate, an 18-point improvement over GPT-5.6 Terra. These are results from that particular test, not an overall probability that a real-world answer will be wrong. OpenAI launch announcement
Coding improved too. On Terminal-Bench 4.0 at maximum effort, the new model solved 50.3% of tasks, versus 21.5% for GPT-5.6 Terra. This makes Terra more credible for a focused bug fix, a small feature, or an investigation with tests you can run. OpenAI still positions Sol and Astra above Terra for the most complex coding work. OpenAI launch announcement · Model selection guide
GPT-6.1 Terra pricing: what will a request cost?
The Felo API model page currently lists these prices per million tokens:
| Felo route | Input | Output | Cached input |
|---|---|---|---|
openai-mix, limited-time 70% off | $0.30 | $1.50 | $0.015 |
| OpenAI standard | $1.00 | $5.00 | $0.05 |
| OpenAI Flex | $0.50 | $2.50 | $0.025 |
| OpenAI Priority | $2.00 | $10.00 | $0.10 |
The promotion is attached to the openai-mix provider route. It is not a blanket discount on every route, and Felo's page does not state an end date. Check the selected provider and current price before making a request. Felo route pricing
Here is a concrete example. Suppose you send 100,000 input tokens and receive 20,000 output tokens, with no cache discount or paid tool calls:
| Model | Input cost | Output cost | Total |
|---|---|---|---|
| GPT-5.6 Terra | $0.20 | $0.24 | $0.44 |
| GPT-6.1 Terra, standard route | $0.10 | $0.10 | $0.20 |
| GPT-6.1 Terra, Felo promotional route | $0.03 | $0.03 | $0.06 |
For that mix, GPT-6.1 Terra costs about 55% less at the standard rate, or about 86% less on Felo's promotional route, compared with GPT-5.6 Terra's published standard rate. Your actual savings will vary with the input-to-output ratio, caching, reasoning tokens, tools, and selected route. GPT-5.6 Terra pricing · Felo GPT-6.1 Terra route pricing
There is one long-context detail worth checking before you send a huge prompt: once input exceeds 272,000 tokens, the higher rate applies to the full request. On Felo's listed standard route, that means $2 per million input tokens and $7.50 per million output tokens. On the promotional route, the corresponding listed rates are $0.60 input and $2.25 output. The 1.05M-token context window tells you what can fit, while the pricing threshold tells you what it may cost. Cache writes and tool calls can add charges as well. Felo model pricing · OpenAI model pricing notes
When should you choose Terra, Sol, or Luna?
Choose GPT-6.1 Terra for a defined task that mixes text with images or application screens, or for concise writing and scoped code fixes. It offers a useful midpoint between capability and price. If a task is ambiguous, spans a large codebase, or needs deep research and planning, start with GPT-6.1 Sol; GPT-6 Astra is OpenAI's most capable option for the hardest work. For narrow classification, triage, or very high-volume automation where cost dominates, GPT-6 Luna may be enough. OpenAI model selection guide · OpenAI model catalog
Price alone is a poor way to pick a model. Run the same representative inputs through two candidates, keep the reasoning effort and evaluation criteria consistent, and inspect the final output. For a visual task, check what the model read from the image. For computer use, verify the saved result. For writing, judge accuracy, useful detail, and voice as well as length. OpenAI recommends this kind of side-by-side evaluation when considering Terra. OpenAI model selection guide
How to try GPT-6.1 Terra on Felo OpenAPI
Felo has a live GPT-6.1 Terra model page with a Playground, pricing, and a request example. Create a Felo API key, add balance if your account needs it, and use the model ID gpt-6.1-terra with Felo's Responses-compatible endpoint:
curl https://openapi.felo.ai/api/v1/responses \
-H "Authorization: Bearer $FELO_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6.1-terra",
"input": "Summarize the main changes in this release note for a nontechnical reader."
}'
For a first test, try a document summary, a product-image question, or a small code issue with a clear expected result. Add a source and ask Terra to cite the exact details it used. If you later build a screen-operating agent, configure and supervise the computer-use tools in your application; the API model by itself only answers the request you send. Felo quick start · OpenAI tool documentation
Frequently asked questions
Is GPT-6.1 Terra cheaper than GPT-5.6 Terra?
Yes, at the published standard token rates. Input drops from $2 to $1 per million tokens, output from $12 to $5, and cached input from $0.20 to $0.05. Felo's limited-time openai-mix route reduces the new model's listed rates further to $0.30 input, $1.50 output, and $0.015 cached input. The amount you actually spend depends on your selected route, token use, long-context pricing, and paid tools. Lower token prices do not mean every completed task will cost a fixed percentage less. Felo promotional pricing · New model standard pricing · Previous model pricing
Does the million-token context mean I should paste in everything?
No. Context is the maximum material a model can accept in a request, not a promise that every detail in a massive prompt will be used equally well. Send the relevant pages, screenshots, or files when you can. Large prompts also cost more, and GPT-6.1 Terra's published pricing changes above 272,000 input tokens. For a recurring workflow, test whether retrieval or prompt caching gives you a better result for the money. OpenAI model specifications
Can Terra generate images, audio, or video?
Its listed native modalities are text and image input, text output. The model can reason about an image, but that is different from returning a generated image. OpenAI lists image generation as a tool the model can call through the Responses API when that tool is available; the separate tool has its own behavior and charges. Audio and video are not listed as native input or output modalities for this model. OpenAI model documentation
Is the Felo API the same as choosing Terra in ChatGPT or Codex?
The underlying model name is the same, but the products have different interfaces, plans, and usage rules. The dollar amounts in this article are API token prices, not a ChatGPT subscription price. OpenAI says GPT-6.1 Terra is also available to eligible ChatGPT Work and Codex users; Felo OpenAPI gives developers a model route they can call with a Felo API key. Check the terms and prices for the product you actually use. OpenAI launch announcement · Felo model page
GPT-6.1 Terra is a notable upgrade because the capability gains come with a lower standard token price. Felo's limited-time 70% discount makes the openai-mix route cheaper still. Test it on work you already know how to check, then compare the result and total cost with your current model.
Try GPT-6.1 Terra on Felo OpenAPI →
This post is also available in 简体中文, 日本語, 한국어, 繁體中文, हिन्दी, Français, العربية, Русский, اردو, Bahasa Indonesia, Deutsch, Tiếng Việt, Türkçe, Italiano, ไทย, Español, বাংলা and Português.