[PowerTokens]

What

Adds PowerTokens as an inference provider. I'm on the PowerTokens team.

PowerTokens is an OpenAI-compatible gateway for China-origin models (GLM, MiniMax, Qwen, DeepSeek). It is aimed at developers outside China. One API key works for every model.

• Endpoint: https://api.powertokens.ai/v1
• Key: $POWERTOKENS_API_KEY
• Type: openai-compat
• Defaults: large glm-5.2, small glm-5-turbo
• Models: glm-5.2, glm-5.3, glm-5-turbo, glm-4.7, MiniMax-M3, MiniMax-M2.7, qwen3-coder-plus, deepseek-v3-2-251201

Follows the existing pattern

• internal/providers/configs/powertokens.json mirrors the other openai-compat configs.
• //go:embed + powerTokensProvider() + providerRegistry entry in internal/providers/providers.go.
• InferenceProviderPowerTokens constant, also added to KnownProviders() in pkg/catwalk/provider.go.

Verified

• go build was not re-run after the rebase onto current main
• go test ./pkg/catwalk/ ./internal/providers/ passed (internal/providers ok)
• Live call matching this config: POST https://api.powertokens.ai/v1/chat/completions with model deepseek-v3-2-251201 (the lowest listed per-token price in this file), max_tokens 16, user message "Reply with the single word hi". HTTP 200, returned model deepseek-v3-2-251201, finish_reason stop, content hi, usage prompt_tokens 10 / completion_tokens 1. This was the OpenAI-compatible request the config points at, not a Crush TUI session.

Sources

• Pricing (USD per 1M tokens): https://powertokens.ai/pricing?utm_source=github&utm_medium=crush&utm_campaign=crush
• Model docs: https://docs.powertokens.ai?utm_source=github&utm_medium=crush&utm_campaign=crush
• The docs list the GLM, MiniMax and DeepSeek models as reasoning models. Per-request effort levels are left out for now, as in the CoralBricks config.
• Context and output limits follow the upstream model cards.

Happy to keep this file up to date when models or prices change. Contact: service@powertokens.ai