# OpenAI API Pricing 2026: GPT-6.1 Sol, Astra and Luna

GPT-6.1 Sol costs $2.00 per million input tokens on OpenAI. Here's what it and every other GPT model costs direct, through Azure, Bedrock, OpenRouter and Requesty, and on SayGM, where every GPT model is 22% below OpenAI's standard rate.

_Source: https://saygm.com/blog/openai-api-pricing_
OpenAI API pricing for GPT-6.1 Sol, its newest model, is $2.00 per million input tokens and $10.00 output. SayGM is the cheapest way to buy it, and every other GPT-6 model, at standard speed: $1.56 and $7.80 for 6.1 Sol, 22% below OpenAI. Requesty and OpenRouter both add a fee on top of OpenAI's price, so 100 million 6.1 Sol tokens costs $281 on SayGM, $360 direct from OpenAI and about $380 through either router.

[OpenAI](https://saygm.com/labs/openai) released the GPT-6 family through September and 6.1 Sol at the end of the month, so most pricing tables online are already out of date. Below is what every current GPT model costs on each route, what OpenAI's speed tiers and long prompts add, and which GPT-6 model is worth paying for.

## Table of Contents

- [OpenAI API pricing at a glance](#openai-api-pricing-at-a-glance)
- [GPT model prices: OpenAI vs SayGM vs Requesty vs OpenRouter](#gpt-model-prices-openai-vs-saygm-vs-requesty-vs-openrouter)
- [GPT-6.1 Sol: why it replaces GPT-6 Sol](#gpt-61-sol-why-it-replaces-gpt-6-sol)
- [GPT-6 pricing: Astra, Sol or Luna](#gpt-6-pricing-astra-sol-or-luna)
- [What long prompts and speed tiers add to the bill](#what-long-prompts-and-speed-tiers-add-to-the-bill)
- [GPT-5.5, GPT-5.4, o3 and o4-mini pricing](#gpt-55-gpt-54-o3-and-o4-mini-pricing)
- [Where to buy GPT models](#where-to-buy-gpt-models)
- [What OpenAI sees when you use SayGM](#what-openai-sees-when-you-use-saygm)
- [OpenAI API cost questions](#openai-api-cost-questions)
- [About SayGM](#about-saygm)

## OpenAI API pricing at a glance

![Best Priced OpenAI Providers in 2026, showing that SayGM is cheapest across all models](https://assets.saygm.com/blog/2026/10/fed5e668-aa67-4ccb-a3a7-0ad58d7c3623-best-priced-openai-access-in-2026-3560x2548.png)

Every GPT model in SayGM's catalog is 22% below OpenAI's standard rate, from GPT-6 Luna at $0.078 / $0.39 to GPT-6 Astra at $7.75 / $38.75. The 100-million-token figures throughout this post assume an 80/20 input-to-output mix.

One exception, and it's a different product: OpenAI's Flex tier charges half its standard rate, which is below SayGM. Flex responses are deliberately slower, and requests can be refused with a 429 error when OpenAI is short on capacity, so it suits evaluations and background jobs rather than anything a user is waiting on. The comparisons in this post are all at standard speed.

## GPT model prices: OpenAI vs SayGM vs Requesty vs OpenRouter

OpenAI's standard rates, as listed on Requesty and OpenRouter, which both pass OpenAI's price through: Requesty with a 5% markup and OpenRouter with a 5.5% card fee on credit. SayGM's rates are from its live catalog. All checked 7 October 2026.

Each cell shows input / output per million tokens, then the cost of 100 million tokens at an 80/20 mix.

| Model | OpenAI direct | SayGM | Requesty | OpenRouter |
|---|---|---|---|---|
| GPT-6.1 Sol | $2.00 / $10.00 · $360 | **$1.56 / $7.80 · $281** | $2.10 / $10.50 · $378 | $2.11 / $10.55 · $380 |
| GPT-6 Astra | $10.00 / $50.00 · $1,800 | **$7.75 / $38.75 · $1,395** | $10.50 / $52.50 · $1,890 | $10.55 / $52.75 · $1,899 |
| GPT-6 Sol | $2.00 / $10.00 · $360 | **$1.55 / $7.75 · $279** | $2.10 / $10.50 · $378 | $2.11 / $10.55 · $380 |
| GPT-6 Luna | $0.10 / $0.50 · $18 | **$0.078 / $0.39 · $14** | $0.11 / $0.53 · $19 | $0.11 / $0.53 · $19 |
| GPT-5.6 Sol | $4.00 / $20.00 · $720 | $3.10 / $15.50 · $558 | $4.20 / $21.00 · $756 | **$2.11 / $10.55 · $380 (promotional)** |
| GPT-5.5 | $5.00 / $30.00 · $1,000 | **$3.88 / $23.25 · $775** | $5.25 / $31.50 · $1,050 | $5.27 / $31.65 · $1,055 |
| GPT-5.4 | $2.50 / $15.00 · $500 | **$1.94 / $11.63 · $388** | $2.62 / $15.75 · $525 | $2.64 / $15.82 · $528 |
| GPT-5.4 mini | $0.75 / $4.50 · $150 | **$0.58 / $3.49 · $116** | $0.79 / $4.73 · $158 | $0.79 / $4.75 · $158 |
| o3 | $2.00 / $8.00 · $320 | **$1.56 / $6.24 · $250** | $2.10 / $8.40 · $336 | $2.11 / $8.44 · $338 |
| o4-mini | $1.10 / $4.40 · $176 | **$0.86 / $3.43 · $137** | $1.16 / $4.62 · $185 | $1.16 / $4.64 · $186 |

OpenRouter is currently running 50% off GPT-5.6 Sol ($2.00 / $10.00 against a regular $4.00 / $20.00), with no end date published, so for that one model it's cheaper than SayGM while the offer lasts.

## GPT-6.1 Sol: why it replaces GPT-6 Sol

[GPT-6.1 Sol](https://saygm.com/models/gpt-6.1-sol) costs the same as GPT-6 Sol, $2.00 and $10.00, and OpenAI pitches it as near-Astra intelligence for coding, computer use and professional work at one-fifth of Astra's price. On OpenAI's own figures it matches GPT-6 Astra on DeepSWE v1.1 and comes within 2.1 points of Astra on OSWorld 2.0.

The pricing difference that matters is cached input. GPT-6.1 Sol bills cached tokens at $0.10 per million, half GPT-6 Sol's $0.20. Agents and coding tools resend the same system prompt and file context on every turn, so most of their input is cached, and that's where a 6.1 Sol bill drops below a 6 Sol bill at the same headline rate. On SayGM, 6.1 Sol's cached input is $0.078.

There's no reason to start new work on GPT-6 Sol.

## GPT-6 pricing: Astra, Sol or Luna

The three GPT-6 models are a 100x price spread, so picking the right one matters more than picking the right route.

**[GPT-6 Astra](https://saygm.com/models/gpt-6-astra)** is OpenAI's most capable model at $10.00 and $50.00. If 6.1 Sol handles the task, Astra costs five times as much for a small gain on OpenAI's own benchmarks.

**GPT-6.1 Sol** is the default for most production work, especially agents and coding.

**[GPT-6 Luna](https://saygm.com/models/gpt-6-luna)** is $0.10 and $0.50, built for fast, high-volume tasks: classification, extraction, routing and short replies. It's the cheapest current GPT model by a wide margin, at $14 per 100 million tokens on SayGM.

## What long prompts and speed tiers add to the bill

Two things push an OpenAI bill above the headline rate.

**Prompts over 272K tokens.** The GPT-6 models accept up to about 1M tokens of context, but past 272K the rate steps up to double the input price and one and a half times the output. For GPT-6.1 Sol that's $4.00 and $15.00 on OpenAI. SayGM's long-context rate is $3.12 and $11.70.

**Speed tiers.** OpenAI sells each model at several speeds. Flex is half the standard rate, for asynchronous work. Standard is the rate everyone quotes. Fast costs about twice as much for lower latency, and GPT-6 Astra also has an ultrafast tier at six times standard. [OpenAI's pricing page](https://openai.com/api/pricing/) also lists a 10% uplift for data residency.

## GPT-5.5, GPT-5.4, o3 and o4-mini pricing

Plenty of production traffic still runs on older models, and they're all in the table above. Two are worth a closer look.

GPT-5.5 at $5.00 and $30.00 now costs nearly three times as much as GPT-6.1 Sol per 100 million tokens, for an older model. If you're still on it, moving is the quickest cut available. GPT-5.6 Sol at $4.00 and $20.00 costs twice as much as GPT-6.1 Sol.

o4-mini pricing is $1.10 and $4.40 on OpenAI, or $0.86 and $3.43 on SayGM. It remains a reasonable cheap reasoning model, though GPT-6 Luna undercuts it heavily for tasks that don't need deep reasoning.

## Where to buy GPT models

GPT models are closed, so every route ends at OpenAI's models. What changes is the price, the extras around it, and who sits in the path. Prices below are for GPT-6.1 Sol per million tokens, input / output.

### OpenAI direct

![OpenAI Home Page](https://assets.saygm.com/blog/2026/10/0bca6832-da91-4d5d-aa72-2e3c3567f22b-openai-home-page-1595x691.png)

$2.00 / $10.00. Buying direct is the only way to get every speed tier: Flex and Batch at half price for work that can wait, Standard, and Fast at about double for lower latency. The trade-off is that you pay the list rate, with a 10% uplift if you need data residency.

### Azure

![Azure AI Home Page](https://assets.saygm.com/blog/2026/10/715c64ce-e334-465f-9e56-e5a9f6dcd8fa-azure-ai-homepage-1618x738.png)

$2.00 / $10.00 on Azure's global endpoint, and $2.20 / $11.00 on its US and EU regional endpoints. Azure suits teams that already buy cloud from Microsoft and want GPT billed and governed alongside the rest of it. You pay at least OpenAI's rate, and more if you need a specific region.

### Amazon Bedrock

![Amazon Bedrock Home Page](https://assets.saygm.com/blog/2026/10/aba8c98e-4f7d-4402-a013-2f8b2d577bcc-amazon-bedrock-homepage-1463x519.png)

$2.20 / $11.00 in us-east-1. Bedrock is the route for AWS-first teams who want GPT next to their other AWS services and under their existing AWS agreement. It's 10% above OpenAI's standard rate.

### OpenRouter

![OpenRouter Home Page](https://assets.saygm.com/blog/2026/10/624eccd9-1365-430f-a265-4b32dfe60178-openrouter-homepage-2960x1584.png)

$2.11 / $10.55 once its 5.5% card fee on credit is added (5% by crypto). OpenRouter passes through OpenAI's own endpoints alongside Azure and Bedrock, and adds routing on top: automatic failover when one endpoint errors, and one key for hundreds of models. It's also running 50% off GPT-5.6 Sol at the moment. We break it down further in our [OpenRouter alternatives comparison](https://saygm.com/blog/top-5-openrouter-alternatives-with-real-pricing).

### Requesty

![Requesty Home Page](https://assets.saygm.com/blog/2026/10/12bb0be3-06de-4056-b8e0-3cb52471d861-requesty-home-page-1247x738.png)

$2.10 / $10.50 with its 5% markup. Requesty lists OpenAI's endpoints and Azure's regional ones, includes EU data residency on every plan, and drops the markup to zero if you bring your own provider keys.

### SayGM

![SayGM Home Page](https://assets.saygm.com/blog/2026/10/d4bcd436-3630-4f12-8140-997f650386e3-saygm-homepage-2892x1552.png)

$1.56 / $7.80, 22% below OpenAI's standard rate on every GPT model, with no subscription and no credit fee. Long prompts over 272K tokens are $3.12 / $11.70 against OpenAI's $4.00 / $15.00. SayGM also carries Claude, Gemini and open-weight models behind the same key, and switching is a base URL change:

```python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.saygm.com/v1",
    api_key=os.environ["GM_API_KEY"],
)

response = client.chat.completions.create(
    model="gpt-6.1-sol",
    messages=[{"role": "user", "content": "Review this pull request for bugs."}],
)
print(response.choices[0].message.content)
```

The [GPT-6.1 Sol model page](https://saygm.com/models/gpt-6.1-sol) shows the live rate.

## What OpenAI sees when you use SayGM

OpenAI still receives the prompt on every route, because it runs the model. On SayGM, requests use [anonymous routing](https://saygm.com/blog/is-claude-private-what-tee-verified-actually-means), so OpenAI gets the request but not who sent it, and the gateway in between runs inside hardware-attested confidential compute.

You can also strip sensitive data before OpenAI sees it. SayGM's guardrails are optional, set per API key, and redact personal data, secrets and credentials inside the gateway's enclave before the request leaves. They're pattern-based, so treat them as one layer of protection rather than a complete one.

## OpenAI API cost questions

### How is OpenAI API cost calculated?

Per token, with separate rates for input, cached input and output. Cached input is a tenth of the standard input rate on most current models, which is why reusing the same prompt prefix across calls lowers the bill.

### What is the cheapest OpenAI model?

GPT-6 Luna, at $0.10 and $0.50 on OpenAI and $0.078 and $0.39 on SayGM.

### Is GPT-6.1 Sol more expensive than GPT-6 Sol?

No. The headline rate is the same, and 6.1 Sol's cached input is half the price.

### Is GPT API pricing cheaper through a gateway?

It depends on the gateway. OpenRouter and Requesty add a fee on top of OpenAI's price, outside promotions like OpenRouter's current 50% off GPT-5.6 Sol. SayGM is 22% below it on every GPT model at standard speed.

## About SayGM

SayGM is a drop-in inference gateway for teams who don't want to just take a company's word that their prompts are private. Route to a confidential model and your prompt is sealed inside hardware-verified execution, where not even SayGM can see it. That's not a policy, it's provable. Swap in your existing OpenAI, Anthropic, or Gemini code and you're covered in minutes, at transparent, published rates with no hidden markup.

Say gm to AI at [saygm.com](https://saygm.com).

[Website](https://saygm.com) | [Twitter](https://x.com/say_gm_) | [Discord](https://discord.com/channels/799672011265015819/1343950080465698836) | [Blog](https://saygm.com/blog) | [Medium](https://medium.com/@say_gm_) | [Docs](https://docs.saygm.com/)