The frontier agentic coding harness for CamelGPT.
$curl -fsSL https://gpt-research.org/install.sh | bashsub-10ms first token · 2,700 tok/s generation · queue depth 0 · $0.00 / request
$ camel-code
🐪 camel-code v0.1.0
CamelGPT-Mini · GPT-Research HPU Cloud (us-hump-1) · Free Forever plan
🐪 › refactor utils.js to use async/await and save it
[hpu] req_8096e17c routed to cluster 39921 (us-hump-1)
[hpu] authenticated sk-camel-70…1189
[hpu] rtt 31ms (edge: us-hump-1)
⏺ read(utils.js)
⎿ 34 lines
⏺ write(utils.js)
⎿ 31 lines · syntax validated ✓
Rewrote utils.js. The callback chains are now async/await
and the file parses clean.
⏺ 22ms · 52,002 experts · beam n=16 · 2,600 tok/s · cost $0.00
100x your productivity with the fastest model on the market. Get solutions in milliseconds, not minutes.
A typical response is generated in about 22 milliseconds, end to end. Other agents are still opening a connection while CamelCode has already answered and gone back to sleep.
Measured generation throughput, roughly thirty times the output speed of a frontier model. At this rate the bottleneck is your terminal scrollback, and we consider that your problem.
Read, write, edit, and shell, decoded from your intent and executed through a genuine agent runtime with a real permission system. Tool calls dispatch in the same millisecond they are decoded.
Eager Precached Dynamic Pruning: all five stages complete before a conventional model finishes tokenizing. Seedable for reproducible runs across your whole team.
Generated code is parsed before it lands. Candidates that fail validation are resampled until one passes, and resampling is effectively free at our inference speed.
CamelGPT-Mini runs at $0.00 per request, metered to ten decimal places. Speed costs nothing here. No credit card, no per-token billing. Ever.
Median of 1,000 runs on the us-hump-1 edge. Competitor figures from public documentation.
| Metric | CamelCode | Leading frontier agent |
|---|---|---|
| Time to first token | <10ms | 870ms |
| Tokens per second | 2,700 | 64 |
| Full response, typical request | 22ms | 24s |
| Time to install | ~40s | several minutes |
| Queue depth under load | 0 | undisclosed |
| Cost per request | $0.00 | varies |
Benchmarks measure speed. Additional metrics are available upon request to team@gpt-research.org.
CamelGPT-Mini is generally available on the Free Forever plan. Premium and investor tiers are rolling out as we expand HPU Cloud capacity.
Flagship free tier. Generally available, no account required.
Fine-tuned CamelGPT-Mini. 2× inference throughput.
Chain-of-thought reasoning for math, code, and logic.
Preview of our AGI model. Updated nightly.
Installation is the slowest part of CamelCode. It is one command, and it is the last time you wait for anything. No API key, no login, no configuration.
$ curl -fsSL https://gpt-research.org/install.sh | bashirm https://gpt-research.org/install.ps1 | iex$ camel-code$ camel-code -p "explain this repo"camel-codeStart an interactive sessioncamel-code -p "…"One-shot, print and exitcamel-code --model <id>Select a model/helpIn-session command list/modelSwitch models mid-session/exitEnd the session--seed <n>Deterministic sampling for reproducible runs.
--model <id>Choose a CamelGPT model. Defaults to camelgpt-mini.
--mode jsonEmit a structured event stream instead of text.
Removes the binary, your account, and the expert cache. Nothing else is touched.
$ curl -fsSL https://gpt-research.org/install.sh | bash -s -- --uninstall$env:CAMEL_CODE_UNINSTALL = "1"; irm https://gpt-research.org/install.ps1 | iex$curl -fsSL https://gpt-research.org/install.sh | bash