One fusion, the strengths of many.
The strengths of multiple models for office work, web development, and complex agent tasks, through one model API.
Cost is the total cost of all tasks in each evaluation (USD). Average time is measured in minutes per task.
* Terminal Bench 4.0 uses the 63-task CPU-only subset. Baselines: GPT-6 Astra (max), Claude Fable 5.1 (max with fallback).
↖ Upper left is better: higher scores at lower cost.
The dashed line traces the Pareto frontier: the best scores at different costs.
From office documents and web development to reasoning through science.
Decipher faint figures in scanned tables and connect text, diagrams, and tables across pages to give answers you can verify.
Explore the evidence and results ↗
Characters, levels, and interaction logic: play two browser games generated by Fusion.
Play the generated games ↗
Explain conductivity in pure metals while respecting the conditions of the question.
Read the research example ↗
Explore how models draw on knowledge across fields in histone lactylation.
Read the research example ↗
Reason about how cell fate changes during lateral inhibition.
Read the research example ↗
Apply the conditions of Raoult’s law to vapor–liquid equilibrium.
Read the research example ↗High-performance models for cost-efficient AI workflows.
High-performance coding, agents, and document analysis, built for cost efficiency.
Cost-efficient coding for software agents.
Individual developers and enterprise teams can sign up on the self-service platform, try a model in the Playground, or create an API key to connect their own applications and agents.
Get started →Fusion Max 2.0 is designed for demanding general-purpose tasks, including office documents, web development, and multi-turn agent work. Fusion Code 1.0 focuses on software development, debugging, and terminal tasks for cost-conscious coding workflows. Compare output quality and total cost on your own tasks.
Compare model specifications →Fusion provides OpenAI-compatible Chat Completions and Responses APIs. Follow the integration documentation to configure the API endpoint, model ID, and key in a supported client. Your application or agent provides file access, browser actions, and tool execution.
Read the API and integration docs →Usage is billed at the selected Fusion model’s input, cached-input, and output rates. Internal model-collaboration tokens are not billed. A task’s cost depends on its calls and token usage; long-context rates apply when input exceeds 200K tokens. Prices have been updated across Max and Code.
See full pricing →Our routing model draws on model-capability knowledge accumulated through extensive benchmarking to choose a suitable model at each step, based on the current task context. At critical steps, Fusion brings in multiple models to analyze and solve the problem from different perspectives, then synthesizes their conclusions.
Explore how fusion works →No. Conventional routing usually selects one model at the request boundary. Fusion can decide during the task which models participate, what role each model takes, and where to spend compute on candidate paths, disagreement review, and critical verification. It still returns one unified result.
Read the whitepaper →Time depends on the task, reasoning effort, and execution path. Across the three evaluations with Max 1.0 results, Max 2.0 reduces time per task by 33.3%–50.0%. Speed and quality tradeoffs against other models vary by task. Review the full results and test your own workflow.
Compare scores, costs, and time →Start with our privacy policy. For data retention, regions, upstream models, dedicated resources, or SLAs, contact the team to confirm options against your business requirements.
Contact the team →
TM