Best fit
Complex implementation and review loops where strong code reasoning matters more than raw speed.
Give Claude Code the change that needs careful judgment.
Claude Code fits subtle implementation, review-heavy changes, and follow-ups where the agent must reason through tradeoffs. ProPR gives those tasks the same planning, isolation, pull request, logging, and recovery path as every other agent.
Approved planContext and scope attached
Opus / Sonnet / Haiku
Isolated worker
GitHub pull requestCommits, logs, and follow-ups
Your credentials
Task history kept
GitHub merge decision
Agent fit and access
Access
Run subscription-backed Claude Code, or direct provider credentials.
Role in ProPR
Claude plans, implements, reviews, or handles PR fine-tuning — the review path never changes.
Pick the Claude that fits the phase.
Model IDs are configured in the web UI and usable everywhere a model can be named: issue labels, plan settings, /switch, and /review. A reasoning level can be set globally, per model, or per issue with a level-* label — from low to max, including Claude Code's adaptive auto mode.
| Model | Context | Issue label | Reach for it when |
|---|---|---|---|
| Claude Fable 5 | 1M | llm-claude-fable | Anthropic's top tier — the hardest planning, architecture, and review work. |
| Claude Opus 4.8 / 4.7 / 4.6 | 1M | llm-claude-opus48 / -opus47 / -opus46 | Deep refactors and careful implementation on large codebases. |
| Claude Sonnet 4.6 | 1M | llm-claude-sonnet46 | A strong default for routine implementation and PR follow-ups. |
| Claude Opus 4.5 / Sonnet 4.5 | 200K | llm-claude-opus45 / -sonnet45 | The earlier series, compatible with all Claude Code versions. |
| Claude Haiku 4.5 | 200K | llm-claude-haiku | Fast, inexpensive passes and small follow-ups. |
Bring your own Anthropic plan — ProPR never marks up tokens.
Claude Code runs on a Claude subscription (cost-capped, no per-token billing) or a pay-as-you-go Anthropic API key. You pay Anthropic directly; ProPR adds no markup and resells nothing.
| Plan | Price | Good for |
|---|---|---|
| Claude Pro | $20 / mo | Entry tier for light, single-developer use. |
| Claude Max 5× | $100 / mo | 5× Pro usage limits — sustained Claude Code work. |
| Claude Max 20× | $200 / mo | 20× Pro usage limits — heavy, multi-task days. |
| Anthropic API | Pay-as-you-go | Direct metered per-token billing when you prefer it. |
Prices as published by Anthropic, July 2026.
Model choice
Pick the Claude model that fits the job and give it a readable alias for the team.Phase routing
Route Claude at planning and review — the phases that reward careful reasoning — and mix in faster models for routine implementation.Access model
Run subscription-backed Claude Code, or direct provider credentials.Usage tracking
Enable Agent Tank and ProPR shows Claude usage around each tracked call.
Setup details live in the docs.
ProPR runs the official Claude Code CLI (@anthropic-ai/claude-code), and the setup specifics — CLI versions, model aliases, and credentials — are covered in the documentation. Inside ProPR, Claude follows the same path as every other agent: planning, isolated implementation, pull request, follow-up, and recovery. See the direct Claude Code comparison for the operating boundary.