[scx-ai] Add SCX.ai pricing and model configs

[bhuvan2134686]

Adds pricing and model configs for SCX.ai, an Australian AI platform serving models over an OpenAI-compatible API (https://api.scx.ai/v1).

Companion gateway PR: Portkey-AI/gateway#1760

Files

• pricing/scx-ai.json — pay-as-you-go rates
• general/scx-ai.json — params, max_tokens ceilings and capability flags

Pricing

Prices are cents per token (USD per million ÷ 10,000):

Model	Input $/M	Cached $/M	Output $/M
GLM-5.2	0.822	0.206	2.888
qwen3.8-max	1.815	0.21	5.4461

Source: SCX publishes its USD rates publicly via LLM Gateway, which lists SCX as a provider — https://llmgateway.io/providers/scx-ai-gp

Flagging openly: scx.ai does not currently publish a first-party per-token price page — its pricing page is quote-based, and the authenticated /v1/models endpoint returns AUD list prices rather than these USD rates.

Model configs

Both models take a max_tokens ceiling of 131072 and are configured as text in/out with tool calling.

For GLM-5.2 that ceiling comes from the endpoint's own validator (max_tokens above 131072 is rejected with "expected a value <= 131072") rather than the published catalogue, which advertises a stale 4096.

qwen3.8-max is a very recent addition and is not yet serving on the public API at time of writing — it is absent from /v1/models and chat completions return "Unsupported model". Its rates and limits are supplied directly by SCX. The upstream Qwen3.8-Max is natively multimodal, but image input on SCX's deployment is unverified, so image is deliberately not declared in type.supported until it is confirmed. Happy to split this model into a follow-up PR if you would rather every entry be live-verifiable at merge time.

Validation

• jq empty passes on both files
• python3 scripts/check_duplicate_keys.py → no duplicate keys
• prettier@2.8.8 --check passes
• Every price round-tripped back to USD/M to confirm formatting did not alter the literals