poe-models

Poe Models Skill

Query Poe's OpenAI-compatible model catalog at https://api.poe.com/v1/models to inspect pricing, context windows, supported features (tools, vision, reasoning effort), and endpoints for any bot.

The endpoint is public — no auth required — but the script will send a Bearer token if one is available (resolution order matches poe-usage: POE_API_KEY env → ~/.config/fir/models.json .providers.Poe.apiKey → ~/.config/fir/auth.json .poe.access/.poe.key).

Usage

# Summary table of all bots
bash "$SKILL_DIR/scripts/models.sh"

# Filter by substring (case-insensitive) on id or display name
bash "$SKILL_DIR/scripts/models.sh" --filter claude
bash "$SKILL_DIR/scripts/models.sh" --filter gpt-5

# Full normalized JSON details for one model
bash "$SKILL_DIR/scripts/models.sh" --id claude-haiku-4.5

# Machine-readable JSON for all (filtered) models
bash "$SKILL_DIR/scripts/models.sh" --filter qwen --json | jq '.[].id'

# Raw upstream response (everything Poe returns)
bash "$SKILL_DIR/scripts/models.sh" --raw

# Cap rows in table mode
bash "$SKILL_DIR/scripts/models.sh" --limit 20

Flag	Meaning
--filter STR	Substring match on id + display_name (case-insensitive)
--id ID	Show full JSON for a single bot by exact id
--json	Emit normalized JSON array for the filtered set
--raw	Dump the upstream response as-is
--limit N	Cap table rows

Output fields

Table columns: ID, CTX (context window, k-rounded), PROMPT and COMPLETION (USD per token — multiply by 1e6 for per-1M cost), TOOLS, VISION, EFFORT (advertises reasoning_effort).

Normalized JSON (--json) per entry:

• id, display_name, owned_by, description
• input_modalities — e.g. ["text", "image"]
• supported_features — e.g. ["tools", "streaming"]
• supported_endpoints — e.g. ["/v1/chat/completions", "/v1/messages"]
• context_length, max_output_tokens
• pricing — Poe's raw object (prompt, completion, input_cache_read, input_cache_write — USD per token, as strings)
• reasoning — { required, supports_reasoning_effort }
• tools, vision — convenience booleans

Notes

• Poe's metadata is incomplete for many third-party bots: context_length of 0 means Poe didn't fill it in. Use --raw and grep the description as a last resort.
• For richer parsing (description-mined context windows, hardcoded overrides, endpoint routing rules), see cmd/generate-models/main.go (fetchPoeModels).
• For point balance and usage history rather than the model catalog, use the poe-usage skill.