---
title: Pricing — Corbenic AI
description: One public index, no negotiation games. A fixed share of the public RunPod on-demand rate for your server type, re-based quarterly.
---
[Skip to content](https://www.corbenic.ai/en/pricing#main-content)
[ Corbenic](https://www.corbenic.ai/?hsLang=en)
- [Home](https://www.corbenic.ai/en/?hsLang=en)
- [Product](https://www.corbenic.ai/en/galahad?hsLang=en)
- [Pricing](https://www.corbenic.ai/en/pricing)
- [Savings check](https://www.corbenic.ai/en/savings-check?hsLang=en)
- [About](https://www.corbenic.ai/en/about?hsLang=en)
[Check your savings](https://www.corbenic.ai/savings-check?hsLang=en)
Pricing
# One rate. Every server. Nobody left out.
Your price depends on one thing: how much computing power you run. Not the server type, not the GPU generation. Work out your bill yourself or [talk to us.](mailto:sales@corbenic.ai)
The rate card
## Your price, per server.
| GPU | TFLOPS | Galahad licence per month |
| --- | --- | --- |
| AMD MI355X (288 GB) | 2500 | $1,423.50 |
| B300 SXM (288 GB) | 2500 | $1,423.50 |
| B200 SXM (180 GB) | 2250 | $1,281.15 |
| AMD MI300X (192 GB) | 1307 | $744.21 |
| AMD MI325X (256 GB) | 1307 | $744.21 |
| H100 SXM (80 GB) | 989 | $563.14 |
| H200 SXM (141 GB) | 989 | $563.14 |
| H100 NVL (94 GB) | 835 | $475.45 |
| H200 NVL (141 GB) | 835 | $475.45 |
| H100 PCIe (80 GB) | 756 | $430.47 |
| RTX PRO 6000 (96 GB) | 504 | $286.98 |
| RTX PRO 6000 WK (96 GB) | 504 | $286.98 |
| RTX PRO 6000 MaxQ (96 GB) | 439 | $249.97 |
| RTX 6000 Ada (48 GB) | 364 | $207.26 |
| L40S (48 GB) | 362 | $206.12 |
| A100 PCIe 80GB | 312 | $177.65 |
| A100 SXM (80 GB) | 312 | $177.65 |
| A100 SXM 40GB | 312 | $177.65 |
| RTX PRO 5000 (48 GB) | 268 | $152.60 |
| RTX 5000 Ada (32 GB) | 261 | $148.61 |
| RTX PRO 4500 (32 GB) | 211 | $120.14 |
| RTX 5090 (32 GB) | 210 | $119.57 |
| L40 (48 GB) | 181 | $103.06 |
| RTX 4090 (24 GB) | 165 | $93.95 |
| RTX PRO 4000 (24 GB) | 156 | $88.83 |
| RTX A6000 (48 GB) | 155 | $88.26 |
| A40 (48 GB) | 150 | $85.41 |
| V100 SXM2 (16 GB) | 125 | $71.18 |
| AMD RX 7900 XTX (24 GB) | 123 | $70.04 |
| AMD Radeon PRO W7900 (48 GB) | 123 | $70.04 |
| L4 (24 GB) | 121 | $68.90 |
| RTX 5080 (16 GB) | 113 | $64.34 |
| V100 PCIe (16 GB) | 112 | $63.77 |
| RTX A5000 (24 GB) | 111 | $63.20 |
| RTX 4000 Ada (20 GB) | 107 | $60.93 |
| RTX 4080 SUPER (16 GB) | 104 | $59.22 |
| RTX 4080 (16 GB) | 97 | $55.23 |
| RTX A4500 (20 GB) | 95 | $54.09 |
| RTX 3090 Ti (24 GB) | 80 | $45.55 |
| RTX 4070 Ti (12 GB) | 80 | $45.55 |
| RTX 4000 Ada SFF (20 GB) | 77 | $43.84 |
| RTX A4000 (16 GB) | 77 | $43.84 |
| RTX 3090 (24 GB) | 71 | $40.43 |
| RTX 3080 Ti (12 GB) | 68 | $38.72 |
| RTX 3080 (10 GB) | 60 | $34.16 |
| RTX 2000 Ada (16 GB) | 48 | $27.33 |
| RTX 3070 (8 GB) | 41 | $23.35 |
| RTX A2000 (12 GB) | 32 | $18.22 |
| Mac Studio M3 Ultra 512GB | 28 | $15.94 |
| Mac Studio M3 Ultra 96GB | 28 | $15.94 |
| MacBook Pro M4 Max 128GB | 17 | $9.68 |
| Mac mini M4 Pro 64GB | 9 | $5.12 |
[Run the savings check](https://www.corbenic.ai/savings-check?hsLang=en)
How we start our partnerships
## Three steps, quickly settled
| Feature | What happens | What it costs you |
| --- | --- | --- |
| Scoping call | A quick check on your fleet and workload under 30 minutes, no obligation. | Nothing. If Galahad will not pay off on your numbers, we say so and part as friends. |
| Measured pilot | We prove it on your own servers, with clear success criteria set upfront. | Nothing. You decide if it worked. |
| Rollout | Scale up server by server, at your own pace. | Per-server licences: daily, weekly, monthly or yearly, your choice. |
## The objections we hear most, answered straight.
Doesn't vLLM already do this?
No. Every other cache remembers by approximating, it keeps a fuzzy, compressed version to go fast and gives up when the memory gets big. We remember exactly, bit for bit, provably identical, no matter how much you give it.
Why can't a bigger company just add this?
This is the hard part everyone else avoids. It is not a feature bolted onto an approximate cache, being exact is a different architecture and it closes the shortcuts the incumbents built their performance on.
Does it work on today's hardware?
Yes. It works today, on today's hardware, and cuts real bills for real companies. No new law and no future breakthrough required.
 Corbenic
[mailto:email@email.com](mailto:email@email.com)
Product
- [Galahad](https://www.corbenic.ai/en/galahad)
- [Galahad GitHub](https://github.com/corbenicai/galahad)
Company
- [Corbenic AI](https://www.corbenic.ai/en/about)
Contact
[sales@corbenic.ai](mailto:sales@corbenic.ai)
© 2026 Corbenic AI, Inc. · Delaware C-Corp
[Privacy](https://www.corbenic.ai/privacy?hsLang=en) [Terms](https://www.corbenic.ai/terms?hsLang=en)
```json
{
"@context" : "https://schema.org",
"@type" : "FAQPage",
"mainEntity" : [ {
"@type" : "Question",
"acceptedAnswer" : {
"@type" : "Answer",
"text" : "No. Every other cache remembers by approximating, it keeps a fuzzy, compressed version to go fast and gives up when the memory gets big. We remember exactly, bit for bit, provably identical, no matter how much you give it."
},
"name" : "Doesn't vLLM already do this?"
}, {
"@type" : "Question",
"acceptedAnswer" : {
"@type" : "Answer",
"text" : "This is the hard part everyone else avoids. It is not a feature bolted onto an approximate cache, being exact is a different architecture and it closes the shortcuts the incumbents built their performance on."
},
"name" : "Why can't a bigger company just add this?"
}, {
"@type" : "Question",
"acceptedAnswer" : {
"@type" : "Answer",
"text" : "Yes. It works today, on today's hardware, and cuts real bills for real companies. No new law and no future breakthrough required."
},
"name" : "Does it work on today's hardware?"
} ]
}
```