Costs & value
"Cheapest" means nothing if it fails. We compare agents by cost per solved task: what you'd actually pay, on average, to get one working result. Numbers come only from measured runs, with the cost source printed next to each.
Cost per solved task (measured)
No measured costs yet. This table fills in automatically once runs with recorded costs are published. Every number will link to its run and cost source. We won't estimate. Submit a run with its cost →
How each agent is billed
Billing models change, so always confirm on the vendor's page. Links go straight to the official pricing.
| Agent | How you pay | Client | Models | |
|---|---|---|---|---|
 Claude CodeAnthropic | Requires a Claude subscription or Anthropic API usage. See the vendor's pricing page for current plans. | Proprietary | Anthropic Claude models | Pricing ↗ |
 Codex CLIOpenAI | Open-source client. Usage is billed through a ChatGPT plan or an OpenAI API key. | Open source (Apache-2.0) | OpenAI models | Pricing ↗ |
 Gemini CLIGoogle | Open-source client. Usage is billed through a Google account tier or a Gemini API key. | Open source (Apache-2.0) | Google Gemini models | Pricing ↗ |
 AiderAider-AI (open-source project) | Free and open source. You pay your chosen model provider directly (bring your own API key). | Open source (Apache-2.0) | Model-agnostic (bring your own) | Pricing ↗ |
 OpenHandsOpenHands | Open source to self-host with your own model keys. A hosted cloud is also offered; see the vendor's site. | Open source (MIT) | Model-agnostic (bring your own) | Pricing ↗ |
 ClineCline | Open-source extension. You pay your chosen model provider (bring your own API key). | Open source (Apache-2.0) | Model-agnostic (bring your own) | Pricing ↗ |
 OpenCodeanomalyco (formerly sst) | Open source. You pay your chosen model provider. See the vendor's site for any hosted options. | Open source (MIT) | Model-agnostic (bring your own) | Pricing ↗ |
Cheapest way to start
Open-source, bring-your-own-key agents (Aider, OpenHands, Cline, OpenCode) cost nothing beyond model usage, and you can point them at cheaper models. Subscription agents trade a flat fee for predictable spend.
Best for general tasks?
For everyday, non-coding work, see general assistants and research & decision-making agents. Task benchmarks for these categories are on the roadmap.