The models that matter, by what developers actually route to.
Frontier labs set the ceiling; cheap, fast models carry the volume. Archymedes Cloud is built to route across both — this is the field as of the latest OpenRouter data.
Model30-day tokensIntelligence index$ in / out per 1MContext
GLM 5.3 FlashZ.ai · open56.9T570.15 / 0.501.3M
Hy4 previewTencent56.1T—0.83 / 2.501M
DeepSeek V4.1 FlashDeepSeek · open51.4T—0.10 / 0.501M
GPT-5.6 LunaOpenAI51.4T—0.20 / 1.201.05M
DeepSeek V4 FlashDeepSeek · open43.5T—0.04 / 0.641.3M
Space Bunny AlphaStealth28.7T———
Nemotron 3 Ultra 550BNVIDIA19.4T———
MiMo V2.6 FlashXiaomi · open9.8T——1M
GPT-6 LunaOpenAI5.6T—0.10 / 0.50—
Claude Opus 5.5Anthropic2.2T57.64 / 20—
Claude Sonnet 5.5Anthropic—56——
Claude Fable 5.1Anthropic—53.4—1M
Qwen3.8 MaxAlibaba—53.42 / 6—
GPT-6 AstraOpenAI—52.7——
GPT-6.1 SolOpenAI—51.8——
Gemini 3.8 FlashGoogle——0.89 blend—
Grok 4.5xAI———500K
Muse Spark 1.3Meta——0.10 blend1M
Mistral Medium 3.5Mistral————
Qwen3.8-27BAlibaba · open———256K
Kimi K2.5Moonshot · open————
Usage = tokens routed through OpenRouter over 30 days; it reflects developer choice, not benchmark rank. “—” means no reliable figure. Sources: OpenRouter rankings, Tokenmaxxing tracker, TeamDay OpenRouter analysis, September 2026 model updates.