Token cost benchmark for an autonomous Code/PR Review agent, across 24 models. Prices as of 25 Oct 2026.
| Model | $/1M in | $/1M out | Cost / outcome | Cost / month* |
|---|---|---|---|---|
| Claude Haiku 5.5 | $0.10 | $0.50 | $0.0230 | $230 |
| GPT-6 Luna | $0.10 | $0.50 | $0.0230 | $230 |
| Llama 4 Maverick | $0.25 | $0.87 | $0.0546 | $546 |
| DeepSeek V4.1 Flash | $0.30 | $1.20 | $0.0667 | $667 |
| Gemini 2.5 Flash | $0.30 | $2.50 | $0.0769 | $769 |
| Mistral Large 3 | $0.50 | $1.50 | $0.107 | $1,073 |
| Gemini 3.8 Flash | $0.75 | $3.75 | $0.173 | $1,726 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $0.230 | $2,302 |
| DeepSeek V4 Pro | $1.32 | $3.96 | $0.283 | $2,833 |
| Mistral Large 4 | $1.36 | $4.18 | $0.293 | $2,926 |
| Gemini 2.5 Pro | $1.25 | $10.00 | $0.317 | $3,170 |
| Grok 4.7 | $2.00 | $6.00 | $0.429 | $4,292 |
| Claude Sonnet 5.5 | $2.00 | $10.00 | $0.460 | $4,604 |
| Claude Sonnet 5 | $2.00 | $10.00 | $0.460 | $4,604 |
| GPT-6.1 Sol | $2.00 | $10.00 | $0.460 | $4,604 |
| Gemini 4 Argon | $2.00 | $10.00 | $0.460 | $4,604 |
| Gemini 3.1 Pro | $2.00 | $12.00 | $0.476 | $4,760 |
| Claude Sonnet 4.6 | $3.00 | $15.00 | $0.691 | $6,906 |
| Claude Opus 5.5 | $4.00 | $20.00 | $0.921 | $9,208 |
| Claude Opus 5 | $5.00 | $25.00 | $1.15 | $11,510 |
| Claude Opus 4.8 | $5.00 | $25.00 | $1.15 | $11,510 |
| Claude Fable 5.1 | $10.00 | $50.00 | $2.30 | $23,020 |
| Claude Fable 5 | $10.00 | $50.00 | $2.30 | $23,020 |
| GPT-6 Astra | $10.00 | $50.00 | $2.30 | $23,020 |
*At 10,000 outcomes per month. Cheapest model highlighted.
The clean-path steps this benchmark prices:
This path runs 9 steps: 4 tool calls, 1 reasoning step and 4 decision points. Tool steps make two model calls each, and the agent re-reads its growing context on every call. That compounding is why one Code/PR Review outcome costs about 30x a single chat message ($0.691 on Claude Sonnet 4.6), not the price of one message.
These figures are modeled estimates, not metered bills. We price a generic, representative Code/PR Review workflow across 24 models using the same cost engine as the live estimator, at each model’s published list price (checked 25 Oct 2026), under documented default assumptions for planning loops, tool calls, memory retrieval, sub-agents and context size. Your own process will differ, so use these as starting points, tune the assumptions in the estimator, and validate against your real usage. Illustrative estimates, not financial advice.
On the clean path with default assumptions, an agent for Code/PR Review costs about $0.0230 to $2.30 per outcome depending on the model, or roughly $230 to $23,020 per month at 10,000 outcomes. The cheapest model here is Claude Haiku 5.5 at $0.0230; the most expensive is GPT-6 Astra at $2.30.
An agent does not make one model call. It plans, calls tools, retrieves context and re-reads its growing working context on every step. For Code/PR Review that adds up to about 30x the cost of a single chat message.
Across the 24 models benchmarked, Claude Haiku 5.5 is cheapest at $0.0230 per outcome and GPT-6 Astra is the most expensive at $2.30. A cheaper model is not always the right choice, but it sets the floor for this workflow.
The biggest levers are prompt caching on the base context, fewer planning loops, smaller tool results, less retrieval, and choosing a cheaper model where quality allows. You can test each lever in the live estimator.
These are modeled estimates, not metered bills. Each figure prices a generic, representative Code/PR Review workflow across 24 models with the same cost engine as the live estimator, at each model's published list price (checked 25 Oct 2026), under documented default assumptions for planning loops, tool calls, memory retrieval, sub-agents and context size. Your own process will differ, so treat these as starting points, tune them in the estimator, and validate against your own usage.