SpaceXAI: Grok 4.7 is 50% off on Ramp Router through Oct 6; public catalog still $2/$6

Core Background & Industry Pain Points

Grok 4.7 lists at $2 / $6 per million tokens, and the whole request jumps to $4 / $12 once the prompt hits 200k tokens. On 2026-10-02 SpaceXAI quoted Ramp Router: try Grok 4.7 at router.com at 50% off through October 6. The question is whether that discount shows up on the invoice.

Architecture Highlights & Internals

Ramp Router is one OpenAI Responses endpoint, https://api.router.com/v1. Display labels are not request IDs; GET /v1/models is authoritative. Routing itself is free through 2026, with $26 in credits for new users; model tokens bill at list price. Strategies include Flex, benchmark routing, fallbacks, and per-key spend caps. Catalog row grok-4.7: reasoning low / medium / high / xhigh, 500k context and 500k max output, no Fast tier.

Authoritative Benchmarks & Measured Scores

No new eval and no Router re-test. From the xAI launch post: CursorBench 4.0 46.3% (Grok 4.6 40.4%, GPT-5.6 Sol Max 41.7%, Fable 5.1 Max 51.8%); Terminal-Bench 4.0 37.6% (20.3% / 37.3% / 57.9%); DeepSWE v1.1 high-effort 71.0%. List price stays $2 / $6 versus Sol $4 / $20 and Fable $10 / $50. The Router catalog still shows $2.00 / $6.00, matching xAI docs. The 50% off is only in the SpaceXAI and Ramp Router posts. Confirm each request cost log; do not model $1 / $3 as the durable rate. Vercel AI Gateway ran a separate 40% promo through 2026-09-27.

Developer Hands-on Guide

Create a key at router.com, confirm grok-4.7 via GET /v1/models, then POST /v1/responses with high or xhigh. The post says the window ends 2026-10-06; docs do not restate the terms. On the direct Grok API (150 rps, 50M TPM, us-east-1 / us-west-2 / us-central-1) this discount does not apply, and prompts at or above 200k tokens bill the higher rate for every token.