Model catalog
Vendor list prices · under BYOK you pay these directly; GateLLM adds no markup.
| Model | Vendor | Input $/1M tok | Output $/1M tok | Planning | Routine | Source |
|---|---|---|---|---|---|---|
| Claude Sonnet 5 | Anthropic | $2 | $10 | 95 | 98 | Source |
| Claude Opus 5 | Anthropic | $5 | $25 | 100 | 100 | Source |
| Claude Fable 5.1 | Anthropic | $10 | $50 | 100 | 100 | Source |
| GPT-6 Astra | OpenAI | $10 | $50 | 99 | 99 | Source |
| GLM-5.3-Flash | Zhipu | $0.15 | $0.50 | 87 | 94 | Source |
| DeepSeek V4.1 Flash | DeepSeek | $0.30 | $1.20 | 88 | 97 | Source |
| Qwen3.8-Max | Alibaba | $2 | $6 | 89 | 97 | Source |
Prices as of 2026-09-15
BYOK: your API keys live in gateway memory only; requests route through the gateway straight to vendor APIs, never leaving your network. GateLLM is licensed per instance / memory — never a per-token markup.
Capability is split into { planning, routine } (not a single score): open models approach parity on routine (classify / extract / translate) and only lag on planning (complex reasoning) — orchestration is precisely about putting open models on the tasks where they're at parity. Scores are normalized ranges from public leaderboards (illustrative, not our own eval).