Avataar AI launches Varya video model at $0.005 per second, 27x cheaper than rivals

At a glance:

• Avataar AI unveiled Varya, an open‑weight video generation model costing $0.005 per second (about 0.48 rupees).
• The price is roughly 27 times lower than competing open‑source video models such as Veo, Kling, Luma and Runway.
• Varya is trained on curated Indian data to render local clothing, food, architecture and festivals accurately.

What Varya is and how it works

Avataar AI, a Bangalore‑based startup, announced Varya as one of India’s first home‑grown video AI models. The model can generate video at roughly $0.005 per second, a cost advantage the founder Sravanth Aluru describes as 27 times cheaper than comparable open‑source offerings. Varya is released as an open‑weight model on India’s AIKosh portal, the government‑run repository for AI models and datasets.

The cost reduction stems from a distillation process. Avataar began with Alibaba’s publicly available Wan 2.2 model and compressed its capabilities into a leaner architecture that runs in four steps instead of the original 50. This yields generation that is ten times faster while consuming a fraction of the compute budget. By contrast, models such as Veo, Kling, Luma and Runway typically charge $0.10 or more per second.

Positioning against global rivals

Varya does not aim to out‑perform frontier models from the US or China on raw quality. ByteDance’s Seedance, Kuaishou’s Kling and Alibaba’s Wan are pushing motion realism and integrated audio generation far beyond Varya’s current limits. Avataar’s pitch is instead focused on scale and affordability for a market of 1.4 billion people, where cost sensitivity outweighs the need for cutting‑edge visual fidelity.

The company acknowledges that while Varya may lag behind in motion realism, its price point opens doors for Indian businesses, educators and public‑service providers who previously could not afford video AI. The hypothesis is that a lower‑cost, culturally aware model can achieve faster adoption than a technically superior but expensive Western alternative.

Cultural specificity as a differentiator

A key selling point of Varya is its cultural grounding. Rather than retrofitting a Western‑trained model, Avataar curated a dataset that emphasizes Indian clothing, cuisine, architecture, festivals and everyday settings. Global models trained predominantly on Western data often produce culturally inaccurate outputs, limiting their usefulness for Indian users.

By embedding Indian visual cues directly into the training set, Varya aims to reduce such mismatches. This focus aligns with the broader Indian AI strategy, which prioritises models that serve the local population at prices the market can absorb, rather than chasing the largest parameter counts.

Funding, ecosystem and the IndiaAI Mission

Avataar AI is one of 12 startups selected for the IndiaAI Mission, a roughly $1.2 billion government initiative that provides subsidised GPU compute in exchange for publicly releasing models. The company has raised $55 million from Peak XV Partners and Tiger Global. Earlier this year, fellow Indian startups Sarvam and BharatGen also launched foundational models under the same programme.

The mission’s goal is to foster sovereign AI capabilities and reduce reliance on foreign infrastructure. Avataar’s shift from e‑commerce video tools to a foundational model reflects a wider trend of Indian startups building AI that is both affordable and tailored to domestic needs.

Outlook and potential impact

At $0.005 per second, Varya tests whether a video model optimised for affordability and cultural relevance can gain traction faster than technically superior, cost‑lier alternatives. If successful, the model could accelerate AI‑driven content creation across Indian media, advertising, education and government services.

The next steps for Avataar will likely involve expanding Varya’s feature set—potentially adding audio synthesis and higher‑resolution outputs—while keeping the cost structure lean. Observers will watch adoption metrics closely, as they could signal a shift in how emerging markets evaluate AI models: price and relevance may soon outweigh raw performance.