Think through the task.
Handle planning, debugging and technical decisions with a model built for reasoning-intensive work.
Quantum Ultra 1 · MCLM
A new class of AI models. Built on a new architecture.
Quantel builds models for coding, reasoning, vision and interactive software engineering—with the context, output range and developer interfaces modern teams need.
1M-token context128K-token outputVision input
MCLM classA new class of AI models
Capability focusReasoning · coding · vision
↓
QSA · new architectureBuilt for intelligent engineering work
↓
Quantum Ultra 1Long-context understanding · native streaming
Built for interactive engineeringResearch claims are scoped to Quantel internal evaluation.
1,000,000-token context128,000-token outputVision capable25 Aug 2026 release
A new model class
Built to carry more of the engineering loop.
Multi-Core Language Models are Quantel's model class for serious software work. The product focus is direct: stronger reasoning, capable coding, visual understanding and long-form output in a developer-ready model.
Quantum capabilities
More room for the work that matters.
Quantum models are designed for repository analysis, tool-using coding workflows, screenshots, technical documents and long-running implementation tasks.
Compare model specifications →Context1,000,000 tokensWork across large repositories and documents.
Output128,000 tokensGenerate long-form code, analysis and plans.
VisionImage input supportedInspect screenshots, diagrams and visual artifacts.
Quantel Synthesis Architecture
A new architecture for intelligent engineering.
QSA is the architecture behind Quantum. It is designed around the demands of modern coding agents and software teams while keeping the developer experience focused on the work.
Move from idea to implementation.
Work across repositories, code changes, tool calls and detailed engineering instructions.
Read the visual context.
Submit screenshots, diagrams and other images alongside your request for visual analysis.
Internal evaluation
Designed for the software-engineering loop.
In Quantel's latest internal cycle, Quantum Ultra 1 was evaluated across more than 2,000 test runs over approximately one month. The figures below are internal results, not an independently audited or official third-party leaderboard score.
Terminal-Bench · internal>92.7%Quantum Ultra 1
Corresponding raw baseline~83%Same internal evaluation framework
Observed difference+9.7 ppAt least +11.7% relative to the ~83% baseline
2,000+ individual test runs≈ one month of evaluation and iteration50–100 output tokens/sec* in production-oriented testing
Developer surface
Use the client your stack already understands.
Quantum supports three native protocol interfaces for the tools developers already use. Choose the request shape that fits your stack, then choose the model that matches the workload.
- OpenAI Responses for Codex and coding agents
- OpenAI Chat Completions for existing SDKs
- Anthropic Messages for Claude-compatible clients
POST /v1/responses{"model": "quantum-ultra","input": "Review this repository","stream": true}
Available model families
Choose the profile that matches your workload.
Quantum Lite 1 is tuned for fast feedback loops. Quantum Ultra 1 is built for reasoning-intensive software engineering. Both models support long context, long output and vision input.
Fast feedback for everyday builds.
Use quantum-lite for quick interaction, rapid code edits and high-volume engineering workflows.
Deep reasoning for complex work.
Use quantum-ultra for coding, reasoning, visual analysis and agentic engineering workloads.
Aura and Avatar.
Future model families are separated from the current catalog until they have a public ID and production availability.
See the roadmap →Transparent by design
Clear about what is public, internal and proprietary.
Quantel publishes model specifications, supported interfaces and the scope of its internal evaluation. The detailed implementation of QSA remains proprietary.
PublicModel class, architecture, specs and interfacesWhat developers can understand and integrate.
InternalEvaluation methodology and reported benchmark figuresClaims are scoped to Quantel's environment.
ProprietaryQSA implementation and optimization detailsCommercial technology is not fully disclosed.
Build with Quantel
Bring a new class of AI models into your engineering workflow.
Join the waitlist for early access to the Quntel model series and developer platform.