GPT-6.1 Terra on Amazon Bedrock: Price vs Sol, Regions, Model ID
GPT-6.1 Terra on Amazon Bedrock costs $1/$5 per 1M tokens, half of Sol. Full pricing, regions, model IDs, feature gaps and who should switch.
OpenAI's GPT-6.1 Terra became generally available on Amazon Bedrock on October 26, 2026. On Bedrock's Global cross-Region inference it lists at $1.00 per million input tokens and $5.00 per million output tokens, half the price of GPT-6.1 Sol. For most everyday jobs (drafting, document Q&A, routine agents) it is the cheaper default. Sol still costs more for a reason: it is the model OpenAI and AWS pitch for harder coding and multi-step agent work.
Below we compare the prices, regions, model IDs and feature gaps between GPT-6.1 Terra, GPT-6.1 Sol and the older GPT-5.6 Terra on Bedrock. Every number comes from AWS's own model cards. We have not run these models ourselves, so we don't make any performance claims of our own.
What is GPT-6.1 Terra on Amazon Bedrock?
GPT-6.1 Terra is the newest model in OpenAI's "Terra" line, which sits in the middle of the family. In AWS's model descriptions, Luna is for fast, high-volume tasks, Terra is the balanced everyday model, Sol is for complex coding and agentic workflows, and GPT-6 Astra is the most capable. AWS's launch announcement says Terra offers "a balance of capability, speed, and cost". It also lists these improvements over GPT-5.6 Terra:
- more detailed visual reasoning (it accepts images as well as text)
- more accurate computer use
- better factuality and more natural writing
The headline benchmark comes from OpenAI and was repeated by AWS. On OpenAI's "Agents' Last Exam" evaluation, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra, performed on par with GPT-6 Astra, and finished tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. These are vendor figures. Independent results will vary by task.
AWS also highlighted the model in its November 2 weekly roundup. That roundup notes that ChatGPT Work and Codex can be configured to use the model through Bedrock.
How much does GPT-6.1 Terra cost on Bedrock?
The prices below are AWS's Standard-tier list prices, in US dollars per 1 million tokens, from the GPT-6.1 Terra model card. Bedrock charges a higher rate on the whole request once the input goes past 272,000 tokens.
| Inference option | Input | Cache write (30 min) | Cache read | Output |
|---|---|---|---|---|
| Global CRIS, ≤272K input | $1.00 | $1.25 | $0.05 | $5.00 |
| In-Region / Geo CRIS, ≤272K | $1.10 | $1.375 | $0.055 | $5.50 |
| Global CRIS, >272K input | $2.00 | $2.50 | $0.10 | $7.50 |
| In-Region / Geo CRIS, >272K | $2.20 | $2.75 | $0.11 | $8.25 |
AWS says In-Region prices "include a 10% fee over OpenAI rates", which is why Global cross-Region inference (CRIS) is the cheapest way to call it. Terra supports only the Standard tier. Priority, Flex and Reserved are all unavailable for this model.
GPT-6.1 Terra vs GPT-6.1 Sol vs GPT-5.6 Terra: which is cheaper?
This is the comparison most people are searching for. It uses the short-context (≤272K input) Global CRIS prices from each model card:
| GPT-6.1 Terra | GPT-6.1 Sol | GPT-5.6 Terra | |
|---|---|---|---|
| Bedrock launch date | Oct 26, 2026 | Sep 29, 2026 | Jul 13, 2026 |
| Input / 1M tokens | $1.00 | $2.00 | $2.00 |
| Output / 1M tokens | $5.00 | $10.00 | $12.00 |
| Cache read / 1M | $0.05 | $0.10 | $0.20 |
| Long-context (>272K) input / output | $2.00 / $7.50 | $4.00 / $15.00 | $4.00 / $18.00 |
| Context window | 1,050,000 | 1M | 1,050,000 |
| Max output tokens | 128,000 | 131,072 | 128,000 |
| Input types | Text, image | Text, image | Text, image |
| Ultrafast tier | No | Yes (6× Standard price) | No |
| Bedrock model ID | openai.gpt-6.1-terra |
openai.gpt-6.1-sol |
openai.gpt-5.6-terra |
An illustrative calculation (our arithmetic, not an AWS quote): take a workload of 10 million input tokens and 2 million output tokens a day, all under 272K per request, on Global CRIS. List prices work out to roughly:
| Model | Daily cost | 30-day cost |
|---|---|---|
| GPT-6.1 Terra | $20 | $600 |
| GPT-6.1 Sol | $40 | $1,200 |
| GPT-5.6 Terra | $44 | $1,320 |
Real bills depend on caching, reasoning tokens and request size. Treat these numbers as an estimate. Anyone still on GPT-5.6 Terra pays twice as much for input and 2.4× as much for output as they would on GPT-6.1 Terra.
Which AWS Regions support GPT-6.1 Terra?
Availability depends on which Bedrock endpoint you call. According to the model card:
bedrock-runtime(AWS's recommended endpoint for new apps): no in-Region invocation. You must call a cross-Region inference profile.- US Geo profile
us.openai.gpt-6.1-terra: callable from us-east-1, us-east-2, us-west-1, us-west-2, ca-central-1 and ca-west-1. Requests stay inside a US-centred set of destination Regions (Canadian sources can also route to their own Region). - Global profile
global.openai.gpt-6.1-terra: callable from most commercial Regions, including Frankfurt, Ireland, London, Paris, Stockholm, Milan, Spain, Zurich, Tokyo, Seoul, Osaka, Singapore, Sydney, Mumbai, Hyderabad, Jakarta, São Paulo, Tel Aviv and Cape Town. The table lists Bahrain (me-south-1) as unsupported.
- US Geo profile
bedrock-mantle(OpenAI-compatible endpoint): in-Region only in us-east-1 (N. Virginia).- GovCloud: not supported for GPT-6.1 Terra. GPT-5.6 Terra, by contrast, does have GovCloud in-Region pricing.
What this means for data residency: if your data must stay inside the EU or India, GPT-6.1 Terra currently has no EU or India geographic profile. Only the Global profile is offered there, and it can route requests worldwide. GPT-5.6 Terra has an India geo profile (in.openai.gpt-5.6-terra). Our explainer on Bedrock in-country inference by Region covers how these profiles work.
What features are missing on GPT-6.1 Terra?
The model card lists several gaps that matter in production:
| Feature | GPT-6.1 Terra (bedrock-runtime) | GPT-6.1 Sol (bedrock-runtime) |
|---|---|---|
| Bedrock Guardrails | Converse API only | Chat Completions, InvokeModel, Converse |
| Explicit prompt caching | Responses API only | Responses, Chat Completions, InvokeModel |
| Implicit prompt caching | Responses, Chat Completions, InvokeModel, Converse | Same |
| Structured outputs (JSON Schema) | Yes | Yes |
| Server-side tool use | No (Yes on bedrock-mantle) | No |
| Count tokens API | No | No |
| Application inference profiles | Converse only | InvokeModel and Converse |
Two more details from the card:
- Converse on
bedrock-runtimerejectsdocumentcontent blocks with aValidationException. To send a file, use the Responses API with aninput_filepart. - Quotas use a 10× output burndown: each output token counts as 10 tokens against your tokens-per-minute limit. A single request can include up to 100 images.
AWS commits to keeping GPT-6.1 Terra available until at least October 26, 2027, with a legacy period of at least six months after that.
How do you start using GPT-6.1 Terra on Bedrock?
- Open the Amazon Bedrock console and create a long-term Bedrock API key.
- Install the OpenAI SDK. AWS's sample code uses Python and
pip install openai. - Point the SDK at Bedrock. For
bedrock-mantlethe base URL ishttps://bedrock-mantle.us-east-1.api.aws/openai/v1with modelopenai.gpt-6.1-terra. Forbedrock-runtimethe base URL ishttps://bedrock-runtime.{region}.amazonaws.com/openai/v1with modelus.openai.gpt-6.1-terraorglobal.openai.gpt-6.1-terra. - On
bedrock-runtime, give your IAM identitybedrock:InvokeModelon the inference profile and on your account's default project (arn:aws:bedrock:{region}:{account-id}:project/default). The model card lists this requirement explicitly.
This is the minimal Responses API call from AWS's model card:
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="us.openai.gpt-6.1-terra",
input="Can you explain the features of Amazon Bedrock?"
)
print(response)
What are developers saying about GPT-6.1 Terra?
The biggest discussion we found is the Hacker News launch thread, which had more than 400 comments. Opinions were split on where Terra fits:
- User frisbee6152 wrote that they had "always found Terra an unhappy medium": if Luna isn't good enough, "the jump to Sol tends to also be worthwhile."
- User dr_kiszonka said that on newer benchmarks "Terra 6.1 is only a hair behind Sol 6.1, hopefully making it a useful Sol replacement for slightly easier tasks."
- User h14h argued the case comes down to "Speed", citing third-party Artificial Analysis results. We have not verified that comparison.
We did not find any Bedrock-specific community reports yet, such as quota or latency experiences on AWS. These are the questions readers are asking: whether Terra on Bedrock gets EU-only routing, whether Priority or Flex tiers will come, and when GovCloud support might follow GPT-5.6 Terra's.
Editor's take
For most teams on Bedrock, GPT-6.1 Terra should now be the default OpenAI model to try first. It is half the price of Sol per token and cheaper than GPT-5.6 Terra on every line, and both OpenAI and AWS say it performs better. Keep Sol for the jobs where it earns its price: long coding sessions and multi-step agents, or anywhere you need the Ultrafast tier. The catches are on the compliance side: no GovCloud, no EU or India geo profile, and Guardrails only through Converse. If you are a regulated buyer, check those rows before you migrate. Price alone shouldn't decide it.
Terra is part of a wider push by Amazon to sell many vendors' models side by side. See our coverage of the Amazon–Anthropic deal and Grok 4.7 on Bedrock. Expect more model news at AWS re:Invent 2026 (November 30–December 4). If you're also watching security tooling, today's explainer on AWS Continuum pricing covers AWS's other big October launch.
FAQ
Is GPT-6.1 Terra available on AWS Bedrock?
Yes. AWS announced general availability on October 26, 2026. It is available through the Bedrock console and APIs (Responses, Chat Completions, Converse and InvokeModel on bedrock-runtime).
How much does GPT-6.1 Terra cost per million tokens?
On Bedrock's Global cross-Region inference: $1.00 input and $5.00 output per 1M tokens for requests up to 272K input tokens. Above that, it's $2.00 and $7.50. In-Region and US Geo pricing is 10% higher.
What is the difference between GPT-6.1 Sol and Terra?
Sol is positioned for complex coding and agentic workflows and costs twice as much ($2/$10 vs $1/$5 on Global CRIS). Sol also offers an Ultrafast tier. Terra is the balanced everyday model. OpenAI says it is more than three times faster than Sol on its Agents' Last Exam evaluation.
What is the GPT-6.1 Terra model ID on Bedrock?
openai.gpt-6.1-terra. On bedrock-runtime you call the inference profile us.openai.gpt-6.1-terra or global.openai.gpt-6.1-terra.
What is the GPT-6.1 Terra context window?
1,050,000 tokens of context and up to 128,000 output tokens, according to AWS's model card.
Is GPT-6.1 Terra available in Europe or GovCloud?
European Regions can call it only through the Global inference profile. AWS lists no EU geographic profile. GovCloud is not supported for GPT-6.1 Terra.
Does GPT-6.1 Terra support images?
Yes, it takes image input (up to 100 images per request on Bedrock). It outputs text only. It doesn't accept audio, speech or video.
Is GPT-6.1 Terra better than GPT-5.6 Terra?
OpenAI says it scored 15 points higher on Agents' Last Exam. On Bedrock it also costs 50% less for input and about 58% less for output. We haven't independently tested either model.
Sources
- AWS What's New: OpenAI GPT-6.1 Terra is now generally available on Amazon Bedrock
- Amazon Bedrock User Guide: GPT-6.1 Terra model card
- Amazon Bedrock User Guide: GPT-6.1 Sol model card
- Amazon Bedrock User Guide: GPT-5.6 Terra model card
- Amazon Bedrock User Guide: OpenAI models
- AWS News Blog: Weekly Roundup, November 2, 2026
- Hacker News: GPT-6.1 Terra discussion
Featured image: Server racks inside an Amazon data center, photo published by Amazon (illustrative of AWS infrastructure; not a specific Bedrock facility). Source