OrcaRouter, an OpenAI-compatible LLM gateway developed by CONTINUUM AI, has listed four frontier models—GPT-6 Luna, GPT-6 Sol, Claude Opus 5.5, and Grok 4.7—in its catalog and disclosed detailed information.
Key points
•Four models listed: GPT-6 Luna, GPT-6 Sol, Claude Opus 5.5, and Grok 4.7 are listed on OrcaRouter.
•Detailed comparison provided: Users can compare four models on one screen, including context/output limits, latency statistics, and pricing.
•Model differences emphasized: Context limits, pricing tiers, and endpoints are set differently for each model.
OrcaRouter sequentially listed the four frontier models on September 21 and 22. Although these four models were developed by different companies, the platform allows access through a single API key and a single endpoint protocol. CONTINUUM AI noted that this represents a practical change, enabling users to compare the four models on the same screen.
The four listed models showed significant differences in context and output limits. GPT-6 Luna has a context limit of 1.05 million tokens, Claude Opus 5.5 has 1.0 million tokens, and Grok 4.7 has 0.5 million tokens. Conversely, the output limit is highest for Grok 4.7 at 450,000 tokens, while Luna, Sol, and Opus 5.5 are recorded at 128,000 tokens.
Differences were also observed in the gateway's performance metrics and cost structure across models. The latency statistics recorded in the catalog are based on a 7-day window. GPT-6 Luna recorded a p50 of 644 milliseconds and a p95 of 7,806 milliseconds. Claude Opus 5.5 recorded a p50 of 4,666 milliseconds and a p95 of 10,000 milliseconds. However, CONTINUUM AI explained that most of the latency items for Grok 4.7 were filled with 0, meaning no measurement was available.
The pricing structure also maintains different boundaries for each model. Claude Opus 5.5 maintains a unit price of $4 per 1 million input tokens and $20 per output token, while GPT-6 Luna applies $0.1 and $0.5 based on a boundary of 272,000 tokens, and Grok 4.7 changes its unit price based on a boundary of 200,000 tokens.
