| Ramain | Codex, Claude Code | |
|---|---|---|
| Tasks completed (of 85) | 73 to 76 | 71 to 75 |
| Cost per task, same model | 1× | 1.7× to 4.2× |
| Built for | Work in business software | Writing and running code |
| Runs on | Any screen your team uses | A terminal and a codebase |
| Pricing | $1,500/month with 150,000 credits; then $0.01 per additional credit | Model usage or seats |
How we ran it
We gave Ramain, Codex and Claude Code the same 85 enterprise tasks and ran each on the same model, so the only difference was the agent. A task passed only if the finished work was right.
Results
Cost per task, same modelRamainCodex
Gemini 3.8 Flash · Ramain vs Codex
$0.16
$0.66
GPT-6 Astra · Ramain vs Codex
$1.04
$2.50
On Fable 5.1, Ramain cost $0.79 per task against Claude Code's $1.36, and completed 75 tasks to its 74.
Why the gap
Codex and Claude Code are excellent coding agents. Business software is a different job: long forms, slow portals and many screens per task. Ramain keeps only the context each step needs, so the same model spends far fewer tokens to reach the same result.