Models

Pullfrog model names are rolling aliases, like a latest tag on npm. "Claude Opus" resolves to a concrete version at runtime (anthropic/claude-opus-5-5), and when a new generation ships the alias moves with it.

Selecting a model

Set the model on the Model card of the repository console, and connect the provider that pays for it on the Providers card. The Model card on the org console sets the default every repository follows until it overrides it.

The card's button opens its editor, which has two tabs:

TabWhat it does
ManualPins one model from any provider, with its reasoning effort
SmartPicks one of a provider's models for each piece of work. See Smart.

The Providers card shows each provider with every credential that pays for it (subscriptions, keyless rules and API keys), or a way to set one up.

  • Models marked Free cost nothing per token but still need an OPENCODE_API_KEY, as free models do on every route.
  • On Pullfrog Router without a card on file, every model but DeepSeek Flash is locked.

Reasoning effort

Under a pinned model, the model menu lists the reasoning effort levels that model publishes, below the models. Pullfrog stores the setting as a position on the model's range, so it keeps its place when you switch models.

Smart sets effort for each run itself. To override effort for a single run, use the --effort flag.

Smart

Smart runs cheap work on a provider's cheapest model and hard work on its frontier model. For pull request reviews, a model Pullfrog pays for reads the diff and picks one of four rungs.

RungWork it gets
MinimalIssue labeling, and reviews with no behavioral change
LightNew-issue enrichment, and small pull requests that can be checked line by line
StandardMentions, fixes and other tasks, and self-contained pull requests with logic to verify
DeepLarge pull requests, and changes that reach across many files

Smart is free on personal accounts, and included in the Organization plan and an organization's trial. On an organization without either, the Smart tab shows a lock. A Smart setting saved before the plan lapsed is kept, and runs use the ladder's Standard rung meanwhile.

Ladders

Each provider maps the rungs to its own models. Under Smart, the Provider menu in the Model editor lists the providers in this table.

ProviderMinimalLightStandardDeep
AnthropicClaude SonnetClaude SonnetClaude OpusClaude Opus
OpenAIGPT LunaGPT TerraGPT SolGPT Astra
GoogleGemini FlashGemini FlashGemini ProGemini Pro
DeepSeekDeepSeek FlashDeepSeek FlashDeepSeek ProDeepSeek Pro
xAIGrok 4.3Grok 4.3GrokGrok
Moonshot AIKimi K2Kimi K2Kimi K3Kimi K3
OpenRouterDeepSeek FlashGPT LunaGPT SolClaude Opus
OpenCode ZenGPT LunaGPT LunaGPT SolClaude Opus
Pullfrog RouterDeepSeek FlashGPT LunaGPT SolClaude Opus

Three cases sit outside the table:

  • Custom (OAI), an OpenAI-compatible endpoint, has no ladder: every run takes the endpoint's own model unless the guidance names others by ID.
  • On Pullfrog Router, Smart runs DeepSeek Flash for everything until a card is on file.
  • A per-run --model flag overrides Smart, and Pullfrog for OSS repositories don't use it.

Guidance for the pick

Under Smart, the Model row holds the guidance for the pick as plain text, starting from which model takes which work. Every model the provider offers is listed under it with its ID.

  • A model the guidance mentions, by its name or its ID, takes the work the guidance gives it, and everything else stays on the ladder.
  • The names Claude Haiku, Haiku and claude-haiku-5-5 all mean the same model.
  • Guidance can also keep work away from a model, or steer by size and risk without naming one.

Write it in your own words:

Haiku is fine for issue labeling and docs-only pull requests. Anything under
src/billing/ goes to Opus whatever its size, and so do large refactors.
Never use Haiku for code.

Reasoning effort follows the size of the work either way.

  • The button at the top right of the text resets it to the default, and changing the provider does too.
  • Lockfiles, snapshots, generated files and binary files are already left out when a diff is weighed.
  • Guidance for the review itself belongs in review instructions.

Pullfrog for OSS

Repositories in Pullfrog for OSS have their model costs covered by Pullfrog, so the picker offers the set we fund:

ModelNotes
DeepSeek FlashRecommended, and the default
GPT LunaOpenAI's cost-efficient tier
MiniMax FlashMiniMax's cost-efficient tier
Muse Spark ContributorMeta's discounted tier. Meta may train on the prompts and completions.

A model outside this set falls back to the default, and the PR footer says so. The exception is a provider key stored with Pullfrog: the run then uses your key and your pick. A key that lives only in GitHub Actions secrets doesn't count, because the server can't see it when it decides. The same rule applies to a per-run --model flag.

Model resolution

When a run starts, Pullfrog takes the first model set in this order:

  1. Per-run flag (--model)
  2. Repository model in the console, or Smart's pick when the repository is set to Smart
  3. Workflow input (model:)
  4. Organization default in the console, or Smart's pick when the default is Smart
  5. Auto-select: the best curated model your keys can reach, preferring each provider's top tier
  6. OpenCode's own default

Specifiers use the models.dev format (anthropic/claude-opus-5-5), and Pullfrog's aliases map onto them.

The model also decides which agent runs:

Model and credentialAgent
An Anthropic model with Anthropic credentialsClaude Code
An OpenAI model with a Codex subscription or an OPENAI_API_KEY, on an organization opted in to the experimental Codex CLI agentCodex
Everything else, and every run billed through Pullfrog Router or the OSS programOpenCode
© 2026 Pullfrog, Inc.

On this page