AI unicorn Sarvam has announced the launch of Sarvam Code, a new AI agent designed for engineering teams, during its Epoch 2026 conference. The platform is positioned as a highly efficient and cost-effective solution that builds sustainable code for long-term projects. Its debut introduces a unique architecture and an innovative pricing model aimed at disrupting the competitive AI coding assistant market.
A Novel Architecture for Reliable Coding
Sarvam Code's architecture breaks down workflows into three specialized agents: a planner, a worker, and a verifier. This multi-agent system is designed to reliably determine when a coding task is truly complete, addressing a common weakness in existing models. Features like recoverable checkpoints and plan steering further enhance its capability to handle complex, long-running assignments without starting over.
Performance and Cost-Effectiveness Claims
During the launch, Sarvam presented compelling performance metrics on the Terminal-Bench 2.1 benchmark. The company claims Sarvam Code can solve a task for approximately $2, a fraction of the cost of competitors like Claude Code and OpenAI's Codex. These initial results position the agent as a powerful and economically viable alternative for software development teams.
The Underlying Technology and Strategic Positioning
The reported benchmark success was achieved using Z.ai's GLM-5.2 model, highlighting Sarvam's strategy of building a superior agent harness. This approach focuses on the system's orchestration capabilities rather than relying solely on a proprietary model from the start. The product's value is further enhanced by its India-hosted infrastructure, catering to regional data sovereignty requirements.
An Innovative Pricing Model
A key differentiator for Sarvam Code is its "pay for completed work" pricing structure. This innovative model ensures clients are only billed for successful outcomes, eliminating costs associated with retries or unsuccessful attempts. While this aligns vendor and customer interests, the specific criteria defining a "completed" task will be crucial for its practical implementation.
Market Entry and Future Outlook
Sarvam Code is being rolled out in an early beta phase, available by invitation on the company's Indus platform. This launch was a highlight of the Epoch 2026 conference, which also featured announcements of major infrastructure upgrades and the appointment of a new advisor from Mistral AI. These developments underscore Sarvam's aggressive strategy to capture a significant share of the enterprise AI market.
The introduction of Sarvam Code marks a noteworthy entry into the AI-powered software development landscape. Its multi-agent architecture, impressive cost-performance claims, and outcome-based pricing present a compelling proposition for engineering teams. As the product moves from a limited beta to wider availability, the industry will be watching closely for independent validation of its capabilities.