Group Multiplier Guide: How to Pick the Most Economical Group

When creating an ApiKey you need to choose a group; actual charges = official price × group multiplier. Currently 9 groups are publicly available (snapshot date 2026-09-27), plus exclusive groups such as image generation (visible only after logging in to the console and enabling the corresponding permissions):

🤖 Claude Groups

• claude code: 0.37x — Claude-Max pool, native Anthropic Messages protocol; no client restriction, any coding agent can connect
• Claude Max: 0.5x — Claude Max $200 pool that has passed KYC: more stable, with a higher cache hit rate; also has no client restriction
• Claude-AWSB: 0.7x — Amazon Bedrock channel, close to full capability, stable and reliable; choose this when you need that route

🎯 GPT Groups

• GPT-Azure: 0.47x — official Microsoft Azure channel, top-tier intelligence; works with any third-party tool such as OpenCode and DSH (DeepSeek Harness), and can serve as the main line for heavy coding
• GPT-Pro: 0.11x — 20X Pro pool at a lower price; image generation is not available in this group

🌍 Gemini Groups

• Gemini-aistudio: 0.28x — enterprise pool, economical line, better value; connecting through tools such as Cline and OpenCode is recommended
• Gemini-Vertex: 0.49x — official Google Vertex channel, stable and reliable; recommended for Gemini flash

🚀 Grok & Domestic Models

• Grok: 0.55x — official Grok channel; provides the Grok series models (including grok-build-0.1)
• 国产模型组: 0.15x — DeepSeek, GLM, Qwen, MiMo, Step and other models; models with a free suffix are currently unlimited and free, with no SLA

Which one to pick?

For Claude, choose the claude code group (0.37x, no client restriction); choose Claude Max (0.5x) for higher stability and cache hit rate; choose Claude-AWSB (0.7x) when you need the Amazon Bedrock route; for the GPT / Codex series, choose GPT-Azure (0.47x), or GPT-Pro (0.11x, no image generation) for a lower price; for Gemini, choose according to budget between Gemini-aistudio (0.28x, economical) and Gemini-Vertex (0.49x, official stable); for the Grok series, choose Grok (0.55x); for DeepSeek, GLM, Qwen, MiMo, Step and other domestic open-source models, choose 国产模型组 (0.15x). Combined with cache-read billing when Anthropic Prompt Caching hits, costs can be reduced even further in long-session scenarios. Groups and multipliers may change — always refer to the Model Plaza and the nexthub.cc console.