Avataar launches Varya, offers ₹0.50-per-second AI video
Avataar.ai launched Varya, a 14-billion-parameter video-generation model it describes as a distilled, India-built alternative to global offerings, at a New Delhi event on June 12, 2026. According to CNBC-TV18, CEO and co-founder Sravanth Aluru said Varya generates video at about Rs 0.50 per second, delivering quality comparable to Alibaba's Wan 2.2 at roughly 27 times lower cost; other coverage puts the broader efficiency gain versus global rivals like Google Veo and ByteDance's Seedance closer to 10-20 times. CNBC-TV18 and Inc42 report that Peak XV backs Avataar and that the startup built Varya with support from the IndiaAI Mission, including subsidized compute. Avataar says a full technical report on the model is forthcoming.
Avataar's bet is that video-generation economics, not raw model scale, will decide adoption in price-sensitive markets: a claimed 27-times cost advantage against Alibaba's Wan 2.2, if it holds up under independent testing, would materially change the unit economics for creators, advertisers, and educators who can't currently justify per-clip costs on frontier models like Veo or Seedance. That efficiency-first positioning, backed by government-subsidized compute, is also a data point in India's broader push to build sovereign AI capability rather than relying on foreign platforms.
What happened
Avataar.ai launched Varya, a 14-billion-parameter video-generation model the company describes as a distilled, India-built model, at a New Delhi event on June 12, 2026, per CNBC-TV18. CNBC-TV18 reports CEO and co-founder Sravanth Aluru said the model generates video at approximately Rs 0.50 per second and compared Varya's quality and cost to Alibaba's Wan 2.2, saying it is nearly 27 times cheaper. CNBC-TV18 and Inc42 report that the startup is backed by Peak XV and built Varya with support from the IndiaAI Mission, including subsidized national compute access.
Technical context
Per the available coverage, Varya uses a distillation approach and runs on 14 billion parameters; neither CNBC-TV18 nor Inc42 has published an architecture paper or independent benchmark trace, though Avataar has said a technical report is coming. CNBC-TV18 reports cost-per-second comparisons with global models: Varya reportedly produces about 211 seconds of video for Rs 100, versus roughly 3 seconds for Google Veo 3.1 and around 7 seconds for ByteDance's Seedance 2.0, figures presented by the company and reported by CNBC-TV18.
Industry context
The AI video market has so far been dominated by large, compute-heavy projects from major cloud and platform companies; Inc42 reports Avataar is betting on an efficiency-first approach instead of competing on scale. Companies that pursue efficiency over raw scale typically aim to lower inference cost and broaden their addressable market among creators and small businesses, though affordability gains only translate into adoption when quality and latency are also competitive and tooling integrations exist.
For practitioners
For ML engineers and product teams, the practical open questions are reproducible quality, latency, and total production cost once orchestration, storage, and delivery are included. The reported Rs 0.50-per-second figure should be treated as a vendor claim until independent benchmarks compare output fidelity, resolution, runtime, and compute requirements against alternatives.
What to watch
Worth tracking: independent benchmark studies on visual fidelity and temporal coherence; published inference-cost breakdowns that include serving infrastructure and postprocessing; creator or publisher case studies showing time-to-create and retention metrics; the release of Avataar's promised technical report; and uptake signals from education, commerce, and creator-platform partners. Broader adoption will also depend on integrations with existing video tooling and verifiable track records on hallucination, safety, and content moderation.
Key Points
- 1Avataar built Varya's 14-billion-parameter model with subsidized compute from the government-backed IndiaAI Mission.
- 2CNBC-TV18 reports Varya can produce about 211 seconds of video for Rs 100, versus roughly 3 seconds from Google's Veo 3.1 at similar spend.
- 3Practitioners should treat Avataar's cost and quality claims as vendor figures until independent benchmarks and its promised technical report arrive.
Scoring Rationale
Notable regional model launch, independently confirmed on core facts (14B parameters, launch date, executive quotes) by a second quality outlet beyond the original CNBC-TV18/Inc42 sourcing. The 27x cost claim against Alibaba's Wan 2.2 is a real practitioner signal for cost-sensitive video workloads but remains an unverified vendor figure pending independent benchmarks.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
