Code Supernova: Some Benchmarks

Code Supernova: an execution-focused code model that runs fast

Code Supernova, a stealth model now available for free through Kilo Code, offers a 200k context window and no rate limits. Systematic tests against frontier models reveal a distinct design trade-off: much faster code generation at the cost of architecture and production hardening. The results suggest a new role in multi-model development workflows rather than a straight replacement for reasoning-focused models.

Testing methodology

Tests ran in Kilo Code across realistic tasks:

• Frontend: build a production-ready landing page
• Backend: implement a SQLite-backed job queue with concurrency handling
• Measurement criteria: speed, code quality, architecture, and edge-case handling

Comparisons included Opus 4.1, Sonnet 4, and GPT-5.

Speed analysis

Measured in Kilo Code, Supernova produced complete code 6–10x faster than GPT-5. That speed supports tight iteration loops where developers can generate an initial approach, feed errors back, get fixes, and pivot architecture multiple times within minutes. The fundamental reason: Supernova behaves as an execution model rather than a planning model — it follows instructions quickly and directly, while models like GPT-5 spend time reasoning about architecture and edge cases.

Frontend test: landing page

Prompt: Build a Postgres hosting landing page with hero, key features, pricing tiers, trust elements, and specified styling.

• Supernova produced a fully functional React landing page with Tailwind CSS in about 17 seconds. Visual output rivaled Sonnet 4 and included helpful UI touches like “Most Popular” badges.
• The generated code was a single ~400-line component with copy-pasted sections and little modularity. The result is workable for prototypes and visual validation but presents maintenance challenges for production or team-based code review.

Backend test: job queue

Prompt: TypeScript queue using better-sqlite3, with optional scheduling via delay timestamps.

• Supernova returned a worker-pool implementation, basic job processing, and a retry counter in ~20 seconds.
• Missing elements included transaction rollbacks, job unlocking on failure, cleanup mechanisms, and robust error propagation.
• GPT-5 took longer (several minutes) but produced a more production-ready design: atomic transactions, visibility timeouts, and clear ack/fail/release separation to avoid race conditions.

Knowledge and training cutoff

Supernova’s training cutoff is September 2024, matching GPT-5 but trailing Sonnet 4 and Opus 4.1 (March 2025). As a result, Supernova may produce working code that relies on older patterns and lacks awareness of recent framework updates such as Next.js 15, React 19, newer TypeScript syntax, and Tailwind v4 classes.

Where Supernova fits

A clear pattern emerged: Supernova excels at fast execution and visual output, but is weaker at planning, defensive patterns, and production-grade architecture. Suitable use cases include:

• UI component generation for initial layouts and mockups
• API testing and integration prototypes (quick clients or webhook handlers)
• Proofs of concept to validate feasibility quickly
• Small feature additions where architecture is already decided
• Static page generation and marketing sites with low maintenance requirements

Less suitable for production systems, team codebases requiring modularity, safety-critical code, or complex state management.

Multi-model workflow

Kilo Code’s model switching enables a practical workflow:

1. Generate multiple UI prototypes rapidly with Supernova.
2. Select the preferred visual approach.
3. Refactor and harden the chosen prototype with a planning-capable model like GPT-5.

This leverages Supernova’s speed for iteration and the other model’s strengths for architecture and robustness.

Getting started

Code Supernova is currently free in Kilo Code with a 200k context window and no rate limits. Fast access paths include:

• Kilo Code model page: https://kilocode.ai/models/supernova
• Install Kilo Code for VS Code: https://marketplace.visualstudio.com/items?itemName=kilocode.kilo-code
• Install Kilo Code for JetBrains IDEs: https://plugins.jetbrains.com/plugin/28350-kilo-code

Switch to the Supernova model from the model selector within Kilo Code. The suggested quick test: request a landing page and compare generation time against a planning-focused model to observe the speed differential.

Original analysis and full write-up: https://blog.kilocode.ai/p/testing-code-supernova-vs-sonnetopus