Plans and pricing
Standard
$20 per month
For individuals and small projects.
Includes:
- 25M tokens per five-hour window
- 2B tokens per month
- Standard priority
Plus
$100 per month
For labs, teams and daily agent work.
Everything in Standard, plus:
- 5x the usage of Standard
- 125M tokens per five-hour window
- 10B tokens per month
- Higher priority
Max
$200 per month
For heavy agent pipelines and production research.
Everything in Plus, plus:
- 20x the usage of Standard
- 500M tokens per five-hour window
- 40B tokens per month
- Highest priority
Pay as you go
For occasional use and bursty workloads. No subscription.
$0.50 per million input tokens $2 per million output tokens
- Billed per token, no monthly fee
- No five-hour window or monthly cap
- Standard priority
Proposed plans. Prices are in U.S. dollars. Plan allowances count input and output tokens together and reset every five hours and every calendar month. Pay as you go is billed per token, with no allowance or cap.
Subsidies
Full and partial subsidies for Californians
California researchers, students, public-sector staff, nonprofits and small businesses could have any plan subsidized in full or in part. Tell us who you are and what you'd run.
Compared
How CalCompute compares
The same $20, $100 and $200 price points as ChatGPT and Claude. CalCompute would publish its allowance in tokens and include the API and coding agents in every plan.
Subscriptions
| Plan | Usage, as published | API | Coding agent |
|---|---|---|---|
| $20 a month | |||
| CalCompute Standard | 25M tokens per five-hour window, 2B per month | Included | Included |
| ChatGPT Plus | “Expanded Codex usage.” “Limits apply.” | Billed separately | Codex |
| Claude Pro | “At least 5x more usage per 5-hour session than Free” | Billed separately | Claude Code |
| $100 a month | |||
| CalCompute Plus | 5x Standard: 125M tokens per window, 10B per month | Included | Included |
| ChatGPT Pro | “5x more usage” than Plus | Billed separately | Codex |
| Claude Max 5x | 5x the usage of Pro per five-hour session | Billed separately | Claude Code |
| $200 a month | |||
| CalCompute Max | 20x Standard: 500M tokens per window, 40B per month | Included | Included |
| ChatGPT Pro, higher tier | No usage figure published | Billed separately | Codex |
| Claude Max 20x | 20x the usage of Pro per five-hour session | Billed separately | Claude Code |
Pay per token
| Model | Input, per million tokens | Output, per million tokens |
|---|---|---|
| CalCompute pay as you go | $0.50 | $2 |
| OpenAI GPT-5.6 Luna | $0.20 | $1.20 |
| Anthropic Haiku 4.5 | $1 | $5 |
| OpenAI GPT-5.6 Terra | $2 | $12 |
| Anthropic Sonnet 5 | $2 | $10 |
| OpenAI GPT-5.6 Sol | $4 | $20 |
| Anthropic Opus 5 | $5 | $25 |
| OpenAI GPT-6 Astra | $10 | $50 |
| Anthropic Fable 5.1 | $10 | $50 |
CalCompute figures are proposed. ChatGPT and Claude figures are quoted from OpenAI's pricing, OpenAI API pricing and Claude pricing as of Sept. 21, 2026, billed monthly, before tax. Both providers list their top plan “from $100” and neither page states a $200 price. Claude publishes a Max 20x tier; OpenAI publishes no usage figure above its $100 tier. CalCompute's pay-as-you-go rate is for the open models it would run. OpenAI and Anthropic rates are for their own models.
Cloud servers
Rent the machine, not the tokens
For training runs, fine-tuning and anything that needs the whole GPU. Proposed rates, billed by the second, the same anywhere in California. No fee for data out, no multi-year commitment, support included.
| Proposed rate | On demand | Flex |
|---|---|---|
| H100 80 GB, per GPU-hour | $3.20 | $1.60 |
| B200, per GPU-hour | $5.40 | $2.70 |
| vCPU, per hour, 4 GB memory included | $0.05 | $0.025 |
| Storage, per GB-month | $0.02 | — |
| Data out to the internet, per GB | $0 | — |
Flex jobs run when the cluster has room and can be paused. Checkpoint your work.
Compared with the private cloud
| Item | CalCompute | AWS | Azure | Google Cloud | Lambda |
|---|---|---|---|---|---|
| H100, per GPU-hour | $3.20 | $6.88 | $12.29 | $11.06 | $3.99 |
| H100 interruptible, per GPU-hour | $1.60 (Flex) | $2.60 (Spot) | $2.27 (Spot) | $6.62 (Spot) | — |
| B200, per GPU-hour | $5.40 | $14.24 | — | — | $6.69 |
| 4 vCPU and 16 GB server, per hour | $0.20 | $0.202 | $0.192 | $0.194 | — |
| Object storage, per GB-month | $0.02 | $0.023 | $0.0208 | $0.020 | — |
| Data out to the internet, per GB | $0 | $0.09 after 100 GB | $0.087 after 100 GB | $0.12 after 1 GB | $0 |
| Lower rate for a commitment | None needed | 1 or 3 years | 1 or 3 years | 1 or 3 years | Contact sales |
CalCompute rates are proposed and set to recover cost, not to undercut. They start from UC Merced's research-computing rate of $0.10 per core-hour with 2 GB of memory, and from the San Diego Supercomputer Center's 2024 purchase of 136 H100 GPUs, 3 PB of storage and two years of operations for $5.3 million. That works out to $2.22 per GPU-hour at full use and $3.20 at 70 percent. AWS, Azure, Google Cloud and Lambda figures are published on-demand list prices for U.S. regions as of Sept. 21, 2026, before tax and discounts, divided per GPU where sold by the eight-GPU server. Spot prices change hourly. Sources: AWS, Azure, Google Cloud and Lambda.
Every plan
Every plan includes
The same platform on every plan. Only the allowance changes.
One endpoint, two dialects
An API endpoint compatible with both the OpenAI and Anthropic APIs. Point your existing SDK at CalCompute and keep your code.
Your coding agent
Claude Code, Codex CLI, OpenCode and any tool that speaks either API.
Chat
Chat in the browser, drawing on the same allowance as the API and your agents.
Compare
Compare plans
| Feature | Standard | Plus | Max | Pay as you go |
|---|---|---|---|---|
| Proposed price | $20/month | $100/month | $200/month | Per token |
| Tokens per five-hour window | 25M | 125M | 500M | No cap |
| Tokens per month | 2B | 10B | 40B | No cap |
| Usage compared with Standard | 1x | 5x | 20x | — |
| Priority | Standard | Higher | Highest | Standard |
| API endpoint | Yes | Yes | Yes | Yes |
| Coding agents | Yes | Yes | Yes | Yes |
| Chat | Yes | Yes | Yes | Yes |
Frequently asked questions
- What counts as a token?
- Input and output tokens across the API, coding agents and chat, added together.
- What happens when I reach the limit?
- Requests would pause until the five-hour window or the month resets. You could move up a plan or switch to pay as you go at any time.
- Who qualifies for a subsidy?
- Californians doing research, study, public-sector, nonprofit or small-business work. Apply and we follow up by email.
- Can I switch plans?
- Yes, up or down, taking effect the next month.
- Which tools work?
- Anything that speaks the OpenAI or Anthropic API: Claude Code, Codex CLI, OpenCode, Cursor and your own code.
- How does this compare with ChatGPT and Claude?
- The price points are the same: $20, $100 and $200 a month. CalCompute would state its allowance in tokens rather than messages or multipliers, and every plan would include the API endpoint and coding-agent support. ChatGPT and Claude bill API use separately.
- Can I rent a whole GPU instead of buying tokens?
- Yes. Cloud servers would be billed by the second per GPU-hour or vCPU-hour, with storage per gigabyte-month and no fee for data out. Request cloud servers and tell us what you'd run.
- What is Flex?
- Half-price capacity that runs when the cluster has room and can be paused. Checkpoint your work and it would be the cheapest way to train.
- Are these the final prices?
- These are proposed plans. Request access now and we'll follow up by email.
Build on California's public cloud
Tell us about your work and what you need to run. We'll follow up by email.