How routing works
Fast Race
Groq + Cerebras + NVIDIA NIM race simultaneously
OpenRouter Race
20+ free models race in parallel — first wins
Sequential Fallback
GitHub → Mistral → SambaNova → Cloudflare
Cache Hit
Repeated prompts served in <10ms from cache
NVIDIA NIM — build.nvidia.com
Free API · OpenAI-compatible · Fast inference
MiniMax M2.7
minimaxai/minimax-m2.7
Params
230B
Context
65K
Main coding agent. Complex software engineering, agentic tool use, Agent Teams, dynamic tool search. Deeply participates in its own evolution.
Kimi K2 Thinking
moonshotai/kimi-k2-thinking
Params
1T MoE
Context
256K
Most capable open-source thinking model. Reasons step-by-step while dynamically invoking tools. 256K context window.
Kimi K2 Instruct
moonshotai/kimi-k2-instruct
Params
1T MoE
Context
131K
State-of-the-art MoE model with 32B activated parameters. Excellent for coding and instruction following.
GLM-4.7
z-ai/glm4_7
Params
7B
Context
32K
Multilingual agentic coding partner. Stronger reasoning, tool use, UI generation, and terminal-based tasks.
Phi-3.5 Mini
microsoft/phi-3_5-mini-instruct
Params
3.8B
Context
128K
Microsoft's compact powerhouse. Fast code generation with 128K context. Perfect for quick tasks.
Phi-3 Medium 128K
microsoft/phi-3-medium-128k-instruct
Params
14B
Context
128K
128K context window for analyzing large codebases, long documents, and complex multi-file tasks.
Mamba Codestral 7B
mistralai/mamba-codestral-7b-v0.1
Params
7B
Context
256K
Mistral's code specialist. Optimized purely for code generation with 256K context.
Gemma 2 27B
google/gemma-2-27b-it
Params
27B
Context
8K
Google's best open model. Used for security analysis, OMEGA plan generation, and recon AI.
Falcon 3 7B
tiiuae/falcon3-7b-instruct
Params
7B
Context
32K
TII's fast lightweight model. Quick responses for simple tasks and chat.
Other Providers
Fallback chain — zero downtime
Groq
Custom silicon — ~500 tok/s
Cerebras
Wafer-scale chip inference
OpenRouter
20+ free models in parallel race
GitHub Models
150 req/day free per token
Mistral
European AI, strong code
AICC (Elite)
Premium models for Elite plan
Cloudflare
10K neurons/day free
SambaNova
High-throughput inference
Plan-based model access
Free
- ✓All NVIDIA NIM free models
- ✓OpenRouter 20+ free models
- ✓Groq + Cerebras
- ✓GitHub Models (150/day)
Starter
- ✓Everything in Free
- ✓Managed API keys
- ✓MiniMax M2.7 (primary)
- ✓Kimi K2 Thinking
Pro
- ✓Everything in Starter
- ✓Web search + Image gen
- ✓Stable Diffusion 3
- ✓Codestral (code)
Elite
- ✓Everything in Pro
- ✓GPT-5.4 Pro
- ✓Claude Opus 4 Thinking
- ✓Gemini 2.5 Flash
- ✓Grok 4.1
Start using all these models today
Free tier includes NVIDIA NIM, OpenRouter free models, Groq, Cerebras, and GitHub Models. No credit card required.