Models & Smart Routing
Orchestrate frontier foundation models through an intelligent cascade router with automatic fallback and zero-training guarantees.
Kernax Auto Waterfall Routing
Different engineering tasks require different model strengths. Kernax integrates an intelligent multi-provider cascade router that dispatches queries to the most cost-efficient, low-latency commercial inference gateway available.
Our routing engine dynamically monitors upstream health, queue latency, and rate limits across a prioritized cascade:
- Waterfall Priority Order: Inference requests prioritize Fireworks AI for bare-metal GPU throughput, seamlessly cascading to Together AI and OpenRouter if failover occurs.
- Guaranteed Non-Training Flag: Every payload dispatched across our gateways programmatically enforces
data_collection: deny, ensuring your code is never logged, cached on disk, or used to fine-tune foundation models. - Stateless In-Memory Transit: Payloads reside strictly in volatile RAM for the duration of the streaming response and are instantly purged upon completion.
Frontier Foundation Models
Kernax offers direct access to the world's most capable programming foundation models, backed by verified context windows and high-output token budgets:
| Model | Context Window | Max Output | Core Competency & Strengths | Recommended Use Case |
|---|---|---|---|---|
| Claude Sonnet 4.6 | 200,000 tokens | 16,384 tokens | Deep multi-file architectural understanding, surgical code refactoring, strict instruction compliance | System design, multi-file code editing, deep PR reviews |
| Claude Opus 4.6 | 200,000 tokens | 16,384 tokens | Maximum reasoning depth, formal logic verification, complex dependency resolution | Mission-critical architecture, security vulnerability audits |
| GPT-5.2 Codex | 256,000 tokens | 16,384 tokens | Frontier autonomous coding, terminal subprocess orchestration, tool execution fidelity | Autonomous agent workflows, CLI scripts, test generation |
| DeepSeek R1 | 128,000 tokens | 16,384 tokens | Chain-of-thought mathematical reasoning, formal algorithm design, race-condition analysis | Algorithmic optimization, compiler logic, concurrency debugging |
| Gemini 3.8 Flash | 1,048,576 tokens | 16,384 tokens | Massive monorepo indexing, ultra-low latency, comprehensive context digestion | Large codebase analysis, full-repo documentation synthesis |
| Kernax Auto (GLM 5.3 Flash) | 1,048,576 tokens | 16,384 tokens | Sub-50ms latency, high throughput, zero-cost execution on Free Tier | Default Free Tier model for daily chat and rapid code generation |
| Kimi k2.5 | 262,144 tokens | 16,384 tokens | Extended-context document synthesis and conversational continuity | Large specification parsing, legacy codebase comprehension |
| Qwen 2.5 Coder 32B | 32,768 tokens | 16,384 tokens | Specialized high-speed Fill-In-the-Middle (FIM) code completions | Sub-millisecond Tab autocomplete engine |
Free Tier Quota & Master Wallet
Kernax provides an accessible pricing model built for individual developers and high-velocity engineering teams:
- Closed Beta Free Tier Quota: Every approved closed beta account receives a complimentary quota of 150 prompts powered by our default 1M-context Kernax Auto model.
- Corporate Master Wallet: Eliminate individual credit card management. One centralized balance funds your entire engineering organization with exact per-token billing for premium flagship models (billed at wholesale gateway cost + transparent margin).
- Zero Seat Waste: Developers who only use the IDE occasionally cost $0.00 in idle seat licenses. You only pay when actual code generation occurs.
Quick-Picker Model Selector
Switch between foundation models at any time directly from the keyboard:
Press Ctrl+/ / Cmd+/ or click the active model badge in the chat composer to open the Model Quick-Picker. You can pin preferred models per workspace or assign specific models to background Mission Control pipelines.