Which API should you use with Crush?

For Crush on one monthly price, the best API is an OpenAI-compatible provider with a fixed bill: add Standard Compute to your crushrc as an openai-compat provider at https://api.stdcmpt.com/v1, from $19 a month with a $20 compute budget. Crush is free and runs on your own keys. Charm's own Hyper service bills Hypercredits at 5 cents each, with 100 free a month or 250 a day on its $20 subscription, and OpenRouter suits light, bursty use.

Guide updated 2026-10-14. Check each provider's current terms below.
Light or occasional use
For occasional sessions, Hyper's 100 free monthly Hypercredits or a pay-as-you-go key on an efficient model will likely cost the least. A local model through Ollama works too; Crush's own docs use it as the first example.
One fixed monthly bill
When Crush is your daily terminal agent, a Standard Compute key keeps the model bill at the plan price, with Claude, GPT and open models behind one provider entry. Requests stop at the monthly compute budget instead of billing overage.

API and subscription options for Crush

OptionPriceWhat you get
Hyper (Charm)$0 or $20/mo, or prepaidCharm's own inference service for Crush, billed in Hypercredits at 5 cents each. Free gets 100 a month; the $20 subscription gets 250 a day; prepaid bundles ($5 for 100) never expire. Source: hyper.charm.land
Standard Compute (us)From $19/moSmart routing sends each request to the cheapest model that can do it, or pin any model by its exact ID. Standard Compute starts at $19/month with a $20 monthly compute budget. Requests stop when that budget is exhausted until renewal or a plan change; there are no automatic overage charges. Requests are not slowed as the budget is used. It is a fixed bill with a finite allowance.
OpenRouterPay-as-you-goChoose models and providers through one API. Usage-based pricing can suit light or bursty work; estimate long agent sessions and review current fees and limits. Source: openrouter.ai
Local model (Ollama / vLLM)Hardware + electricityNo hosted per-token bill, but hardware, context capacity, throughput and maintenance still limit what you can run. Test task quality on your own hardware. Source: docs.ollama.com

Prices and allowances can change. Crush's crushrc format and custom provider types come from its README and config docs; Hyper's prices from hyper.charm.land, both read on October 14, 2026. Crush: custom providers · Crush config reference · Hyper pricing. Spot an error? Tell us.

Add Standard Compute to your crushrc

~/.config/crush/crushrc
provider add standardcompute --name "Standard Compute" --type openai-compat \
  --base-url "https://api.stdcmpt.com/v1" \
  --api-key "$STANDARDCOMPUTE_API_KEY"

model add standardcompute/standardcompute --name "Standard Compute (routed)" \
  --context-window 200000 --default-max-tokens 16000
model add standardcompute/anthropic/claude-sonnet-5.5 --name "Claude Sonnet 5.5" \
  --context-window 200000 --default-max-tokens 16000

Export STANDARDCOMPUTE_API_KEY, start crush and pick the model in its model picker. Use openai-compat, not openai: Crush reserves openai for OpenAI's own API.

Crush questions

Does Crush still use crush.json for custom providers?
It still reads it, but the JSON format is deprecated. Crush now configures itself with a crushrc, a Bash file with Crush builtins such as provider add and model add, looked up in ./.crushrc, ./crushrc and then ~/.config/crush/crushrc. Because it is Bash, the key can come from an environment variable ($STANDARDCOMPUTE_API_KEY) instead of sitting in the file.
Can Crush handle model IDs with a slash, like anthropic/claude-sonnet-5.5?
Yes. In model add, only the first slash separates the provider name from the model ID, so standardcompute/anthropic/claude-sonnet-5.5 registers anthropic/claude-sonnet-5.5 on the standardcompute provider. That is how you pin one model; standardcompute/standardcompute lets Standard Compute's routing pick per request.

Try one task before choosing a plan

  1. Check the API format and model configuration supported by your installed Crush version.
  2. Run a small task on a clean branch and review the result, elapsed time and compute used.
  3. Estimate your actual monthly workload before selecting a larger allowance.
Get your API keyCrush setup guide →Compare compute budgets →

Record your own results with the workload CSV and report template. These are blank templates, not published benchmark results.

Disclosure: this guide is maintained by Standard Compute, one of the options above. Choose based on task quality, supported features and total cost for your workload. Light usage may cost less on a usage-based API; sustained workloads need enough allowance on any subscription.

All flat-rate plans compared →The cheapest-tokens math →Plan compute for your team →