Free 100M tokens for new users. $0.04 per 1M tokens for input tokens.
Per-execution pricing: one fee per execution that returns proof, charged only when proof comes back. Failed or cancelled runs cost nothing. Planning and approvals are free.
Key features
384K token context window
Multimodal input support
100K tokens processed in less than a second
Structured output for direct code use
Subquadratic Sparse Attention algorithm for efficient compute
$0.04 per 1M tokens pricing
Customer content not used for model training
Data retention limited to 60 days
Boundary approval system for controlled execution
Checkpoint-based resumable execution
Verifiable proof of completion
Composable proof for multi-agent workflows
Per-execution pricing model
Integration with multiple agent frameworks
What makes it different
First long-context decision model purpose-built for enterprise work
Frontier-level decision-making performance at 5-10x lower cost than frontier LLMs
Tied for #1 on Artificial Analysis Decision Index
Processes entire codebase or data room in single context window
Speeds agent reasoning from minutes to seconds
Pay only for verified results, not for attempts or planning
Human-approved boundaries ensure agent actions stay within limits
Interruptible checkpoints allow pausing and resuming executions
Proof is built into results for auditability and composability
Subquadratic
Kairos