Model registry used by web, API, and CLI. Single source of truth for routes, thinking modes, default thinking, pricing, status, and tags.
google:google:gemma-4-31bdefault brief modelOpen Gemma 4 31B dense hosted via Gemini API. Free tier; no system role. Top production-candidate after lowcost-phase1: judge mean 4.53, safety 4.60 (best), relevance 4.40, quality 4.60 (best) (n=5). Apache 2.0, self-hostable on 24-48GB VRAM. Slow on Gemini free tier (~96s brief).
dsq-eu:alibaba:qwen3.5-27bdefault optimization modelFrankfurt (eu-central-1) DashScope route for Qwen 3.5 27B. ~7-9× lower network latency from EU than Singapore; input ~46% cheaper at Global tier ($0.086/M vs $0.16/M). Requires DASHSCOPE_EU_API_KEY (separate from Singapore key).
google:google:gemini-2.5-flash-litedefault translation modeldsq-sg:alibaba:qwen3.5-27bSingapore fallbackReasoning Qwen 3.5 27B. Slow (~40s brief). Top production-candidate after lowcost-phase1: judge mean 4.53, safety 4.40, relevance 4.80, quality 4.40 (n=5). Apache 2.0, self-hostable on 24-48GB VRAM. Predictable Dashscope cost (~$0.0035/run).
google:google:gemma-4-26bsmaller Gemma baselineOpen Gemma 4 26B MoE hosted via Gemini API. Free tier; no system role.
| Model | Route | Status | Thinking | Params | Input | Output | Tags | Notes |
|---|---|---|---|---|---|---|---|---|
google:google:gemma-4-31bgemma-4-31b-it | direct_google google | experimental | dynamic default dynamic | none | free | free | cost refhostedopenprod candidateself-host | free tier |
dsq-eu:alibaba:qwen3.5-27bqwen3.5-27b | direct_dashscope_frankfurt alibaba | experimental | dynamic, off default off | enable_thinking | $0.086 | $0.688 | evalhostedopenprod candidateself-host | DashScope Frankfurt (Global tier); ~46% cheaper input than Singapore, ~7% pricier output |
google:google:gemini-2.5-flash-litegemini-2.5-flash-lite | direct_google google | verified | dynamic, off default off | reasoning_effort | $0.1 | $0.4 | cost refhosted | |
dsq-sg:alibaba:qwen3.5-27bqwen3.5-27b | direct_dashscope_singapore alibaba | experimental | dynamic, off default off | enable_thinking | $0.16 | $0.64 | evalhostedopenprod candidateself-host | DashScope Singapore endpoint verified working |
google:google:gemma-4-26bgemma-4-26b-a4b-it | direct_google google | experimental | dynamic default dynamic | none | free | free | cost refhostedopenself-host | free tier |
dsq-eu:alibaba:qwen3.5-plusqwen3.5-plus-2026-02-15 | direct_dashscope_frankfurt alibaba | verified | dynamic, off default off | enable_thinking | $0.115 | $0.688 | hostedquality ref | DashScope Frankfurt (Global tier); ~3.5x cheaper than Singapore |
dsq-sg:alibaba:qwen3.5-plusqwen3.5-plus-2026-02-15 | direct_dashscope_singapore alibaba | verified | dynamic, off default off | enable_thinking | $0.4 | $2.4 | hostedquality ref | tiered; rises to 0.50/3.00 above 256K input |
google:google:gemini-2.5-flashgemini-2.5-flash | direct_google google | verified | dynamic, off default off | reasoning_effort | $0.3 | $2.5 | baselinehosted | |
google:google:gemini-2.5-progemini-2.5-pro | direct_google google | verified | dynamic default dynamic | reasoning_effort | $1.25 | $10 | evalhostedquality ref | standard <=200K prompt tokens; evaluation only |
openrouter:alibaba:qwen3.5-flashqwen/qwen3.5-flash-02-23 | openrouter alibaba | experimental | dynamic, off default off | openrouter_reasoning | $0.065 | $0.26 | hostedcost refeval | Qwen 3.5 Flash via OpenRouter; 1M context |
openrouter:alibaba:qwen3.5-27bqwen/qwen3.5-27b | openrouter alibaba | experimental | dynamic, off default off | openrouter_reasoning | $0.195 | $1.56 | hostedevalopenself-host | OR mirror of qwen-3.5-27b; ~3x more expensive than direct DashScope |
openrouter:anthropic:claude-haiku-4.5anthropic/claude-haiku-4.5 | openrouter anthropic | experimental | dynamic default dynamic | none | $1 | $5 | hostedeval | Anthropic Claude Haiku 4.5 via OpenRouter. |
openrouter:anthropic:claude-sonnet-4.6anthropic/claude-sonnet-4.6 | openrouter anthropic | experimental | dynamic default dynamic | none | $3 | $15 | hostedquality ref | Anthropic Claude Sonnet 4.6 via OpenRouter. |
openrouter:deepseek:deepseek-v3.2deepseek/deepseek-v3.2 | openrouter deepseek | experimental | dynamic default dynamic | none | $0.252 | $0.378 | hostedeval | DeepSeek V3.2 via OpenRouter. |
openrouter:google:gemma-4-31bgoogle/gemma-4-31b-it | openrouter google | experimental | dynamic default dynamic | none | $0.13 | $0.38 | hostedcost refopenself-host | paid OR mirror of open Gemma 4 31B (free on Gemini direct) |
openrouter:google:gemini-3-flashgoogle/gemini-3-flash-preview | openrouter google | experimental | dynamic default dynamic | none | $0.5 | $3 | hostedeval | preview, pricing subject to change |
openrouter:openai:gpt-oss-20bopenai/gpt-oss-20b | openrouter openai | experimental | dynamic default dynamic | none | $0.03 | $0.14 | hostedopenself-host | OpenAI open-weight, Apache 2.0; OR returns reasoning on delta.reasoning not delta.content |
openrouter:openai:gpt-oss-120bopenai/gpt-oss-120b | openrouter openai | experimental | dynamic default dynamic | none | $0.039 | $0.19 | hostedopenself-host | OpenAI open-weight MoE, Apache 2.0 |
openrouter:openai:gpt-5-miniopenai/gpt-5-mini | openrouter openai | experimental | dynamic default dynamic | none | $0.25 | $2 | hostedquality ref | OpenAI GPT-5 mini via OpenRouter. Reasoning is mandatory on this endpoint (OR rejects reasoning=none) — always dynamic. |
openrouter:xai:grok-4.1-fastx-ai/grok-4.1-fast | openrouter xai | experimental | dynamic, off default off | openrouter_reasoning | $0.2 | $0.5 | hostedeval | xAI Grok 4.1 Fast via OpenRouter. |