fix: Perplexity Agent catalog is out of sync with GET /v1/models

[github-actions]

Summary

Syncs the Perplexity Agent catalog with the first-party Agent API models page and documented token pricing. Callable IDs now match the provider/model values Perplexity publishes (including perplexity/… namespaces for hosted open models). Stale IDs that no longer appear on the docs page are marked deprecated. Missing frontier models (Claude 5 / Fable, GPT-5.6 Sol/Terra/Luna, Gemini 3.x Flash family, Grok 4.5/4.3/4.20, GLM-5.2, Nemotron Ultra/Lightning) are added as override-only base_model entries with verified USD/MTok costs.

Changes

Added (callable IDs + pricing from docs)

ID	Cost (input/output/cache $/1M)	Notes
anthropic/claude-opus-5	5 / 25 / 0.50	effort L/M/H/xH/max
anthropic/claude-opus-4-8	5 / 25 / 0.50	same
anthropic/claude-sonnet-5	2 / 10 / 0.20	same
anthropic/claude-fable-5	10 / 50 / 1.00	same
openai/gpt-5.6-sol	5→10 / 30→45 / 0.50	tier @ 272k
openai/gpt-5.6-terra	2→4 / 12→18 / 0.2→0.4	tier @ 272k
openai/gpt-5.6-luna	0.2→0.4 / 1.2→1.8 / 0.02→0.04	tier @ 272k
openai/gpt-5.4-mini	0.75 / 4.50 / 0.075
openai/gpt-5.4-nano	0.20 / 1.25 / 0.02
openai/gpt-5	1.25 / 10 / 0.125
google/gemini-3.1-flash-lite	0.25 / 1.50 / 0.025
google/gemini-3.5-flash	1.50 / 9.00 / 0.15
google/gemini-3.5-flash-lite	0.30 / 2.50 / 0.03
google/gemini-3.6-flash	1.50 / 7.50 / 0.15
google/gemini-3.7-flash	0.375 / 1.875 / 0.0375
xai/grok-4.5	2→4 / 6→12 / 0.3→0.6	tier @ 200k
xai/grok-4.3	1.25→2.5 / 2.5→5 / 0.20	tier @ 200k
xai/grok-4.20-reasoning	same as 4.3	reasoning_options = []
xai/grok-4.20-non-reasoning	same	non-reasoner
xai/grok-4.20-multi-agent	same	effort L/M/H/xH
perplexity/deepseek-v4-flash-0731	0.13 / 0.26 / 0.028	effort low/high/max
perplexity/glm-5.2	1.40 / 4.40 / 0.26	effort high/max
perplexity/kimi-k3	3 / 15 / 0.30	effort minimal…max
perplexity/kimi-k2.7-code	0.95 / 4 / 0.19	[]
perplexity/nemotron-3-ultra-550b-a55b	0.25 / 2.50 / 0.25
perplexity/nemotron-3.5-lightning-30b-a3b	0.0115 / 0.17 / 0.00115

Lab metadata added for Grok 4.20 aliases

• models/xai/grok-4.20-reasoning.toml
• models/xai/grok-4.20-non-reasoning.toml
• models/xai/grok-4.20-multi-agent.toml

Namespace / lifecycle

Old ID	Action
deepseek/deepseek-v4-flash-0731	status = "deprecated" → use perplexity/…
moonshot-ai/kimi-k3, moonshot-ai/kimi-k2.7-code	deprecated → perplexity/…
nvidia/nemotron-3-super-120b-a12b	deprecated (not on docs list)
google/gemini-2.5-flash, google/gemini-2.5-pro	deprecated
xai/grok-4-1-fast-non-reasoning	deprecated

Pricing corrections on existing live IDs

• openai/gpt-5.4, openai/gpt-5.5: added 272k context tiers
• google/gemini-3.1-pro-preview, google/gemini-3-flash-preview: aligned to docs rates/cache
• Existing Claude/GPT/xAI/Gemini entries converted to override-only base_model where lab metadata already exists

Evidence

Claim	Source	What it establishes
Full Agent API model ID set + $/1M token rates (including tiers and cache)	https://docs.perplexity.ai/docs/agent-api/models	Authoritative catalog and pricing used for all adds/updates
List-models endpoint format and example IDs	https://docs.perplexity.ai/api-reference/models-get	Confirms provider/model IDs and discovery endpoint
Lab capability inheritance (limits, modalities, reasoning flags)	Existing models/<lab>/… entries for Anthropic, OpenAI, Google, xAI, DeepSeek, Moonshot, Zhipu, NVIDIA	Provider files stay override-only
Grok 4.20 split IDs (reasoning / non-reasoning / multi-agent)	First-party xAI provider peers + OpenRouter multi-agent peer	Reasoning flags and effort sets
GLM-5.2 effort high/max	providers/zhipuai/models/glm-5.2.toml	Relay copies lab effective effort set
Kimi K3 effort list	Perplexity docs (minimal/low/medium/high/xhigh/max)	Matches prior moonshot-ai Perplexity Agent entry

Validation

• Cross-checked every new/changed cost against the Agent API models page PRICING data (2026-08-27).
• Confirmed each new base_model target exists under models/.
• Did not run bun validate or live GET /v1/models (no shell / no API key in this workflow).

Review notes

• Issue’s authenticated GET /v1/models listed a few IDs not on the public docs pricing table (openai/gpt-5-nano, openai/gpt-5.3-codex, perplexity/deepseek-v4-pro-0813, perplexity/glm-5.3). Those were not added—no verified public token prices.
• Stale IDs are deprecated, not deleted, so consumers can migrate.
• Priority/Fast 2× pricing (service_tier: "priority") for Luna/Sol is documented but not modeled as experimental.modes (same as prior Perplexity Agent entries).
• GLM-5.2 cache-read uses the PRICING calculator value $0.26 (page table elsewhere showed $0.14).
• Nemotron entries use the same effort list as the previous Perplexity Agent Nemotron Super entry on this host (low…max), not NVIDIA first-party toggle-only.
• Embedding models called out in the issue remain out of scope (Seperate Embedding Category #572).

Closes #5630

Automated by the issue fixer: https://github.com/anomalyco/models.dev/actions/runs/33050764231