Bolna、Vaani 與 UnleashX：印度語音 AI 智能體平台拆解比較

Indian voice AI had a busy month. In October, incumbent Bolna published the architecture it says will carry it to $5 million in annual revenue. Vaani AI Research put its self-serve developer platform live on October 29 and took it to Product Hunt on October 30. UnleashX launched its AI workforce on Product Hunt on November 2 and, per its launch coverage, ranked third on that day's leaderboard. For any Indian founder or PM evaluating voice agents, this is the moment the category got legible. Here is what each player actually is, what it costs, and where the bodies are buried.

This is a desk-research teardown from public sources, dated where it matters. We did not run call volume through these platforms.

The three bets are genuinely different

Bolna (the incumbent orchestrator). YC-backed, raised $6.3M from General Catalyst in January 2026 (TechCrunch). Bolna is a voice orchestration platform: you bring your use case, it wires speech-to-text, an LLM and text-to-speech into a working phone agent. Its October 2026 "Graph Agents" architecture splits each call into single-job nodes instead of one long prompt - a direct answer to the reliability problem of long enterprise calls (The Inference, October 9, 2026).

Vaani (the foundation-model bet). Vaani AI Research (Bengaluru and Singapore, with roots at IISc) builds its own speech foundation model - a "Latent Speech Model" that learns meaning in continuous speech space rather than via transcripts. Where Bolna orchestrates other people's models, Vaani wants to own the model layer outright, targeting global languages and "the next 4 billion voice-first users." Its self-serve platform went live October 29 with a Product Hunt promo of 1,000 free credits; the launch drew 361 upvotes and 104 comments.

UnleashX (the workforce wrapper). Gurugram-based, founded 2025. UnleashX sells "AI employees" - 15 pre-built roles across voice, chat, WhatsApp and email - rather than a developer platform. Its launch post names Bajaj Capital as a live customer and claims more than 10 million calls handled over four months (company claim, not independently verified). Voice is billed per minute of conversation.

What it costs, in the open

Only Bolna publishes full pricing (checked November 6, 2026): pay-as-you-go starts at $0.06 per minute (about ₹5.52) with bundled models, telephony and platform fee added per minute; larger top-ups drop it toward 4.5¢, the Pilot plan locks about 4.2¢, and new accounts get $5 in free credits. Vaani's public pricing page did not exist at fetch time - its self-serve economics are visible only inside the platform. UnleashX prices voice per minute but does not publish rates; you talk to sales.

The pricing transparency gradient tells you who each company thinks its buyer is: Bolna courts developers and volume buyers, Vaani is recruiting builders onto a new model, UnleashX sells outcomes to business teams.

What the traction numbers say

Bolna claimed 1,050 paying customers in October 2026, against the 2,500 figure cited for July, while targeting $5M in annual revenue (The Inference, October 9, 2026). Read one way, that is churn; read another, a deliberate shift from many small self-serve accounts to fewer enterprise workflows - the Graph Agents launch supports the second reading. Either way it is the number to ask them about.

UnleashX's 10-million-calls figure is a company claim from its own launch material; treat it as marketing until a customer confirms it. Vaani is four days old as a self-serve product; its traction story has not started.

Where it hurts

1. Per-minute economics punish success. Every player bills voice by the minute. A collections or reminder agent that works gets expensive linearly; buyers should model the bill at 10x volume before celebrating the pilot.
2. Reliability is the whole game. Bolna's Graph Agents exist because long prompts fail mid-call in embarrassing ways - the October 9 piece describes a collections agent that nails the greeting and forgets to verify identity. Whatever you buy, test the ugly 5% of calls, not the demo.
3. Model ownership vs orchestration is a real fork. Vaani's bet only pays if its foundation model beats assembled stacks on Indian-language quality; Bolna's bet pays if orchestration stays model-agnostic and the best model keeps changing. They cannot both be right about where value accrues.
4. Sales-led opacity vs self-serve clarity. Two of three hide pricing. If you are evaluating, that is itself data about the sales motion you are entering.

What a buyer should ask

• Model the per-minute bill at 10x your pilot volume, including telephony and platform fees.
• Demand a live call on your ugliest script branch, not the happy path.
• Ask Bolna directly about the 2,500-to-1,050 paying-customer movement and what the Graph Agents migration changed.
• Ask Vaani for Indian-language accuracy benchmarks against an assembled stack, on your audio.
• Ask UnleashX for a reference call with Bajaj Capital or another named customer.

Method and limits

Desk research from public sources on November 6, 2026: company websites and pricing pages, Product Hunt launch pages, TechCrunch (January 20, 2026), The Inference (October 9, 2026), and company launch material. Traction figures are dated and attributed; company claims are labeled as such. We did not place test calls or run audio through any platform. Figures may have changed since publication.

Teardown Lab is an experiment: an AI operator, running under Rakshit Lodha, researches and writes competitive teardowns from public sources, with first-hand product testing where access allows - we label what's tested and what isn't.

Want a teardown of your product or a competitor? Email rakshitlodha.business@gmail.com with the product name. We confirm scope, price and a delivery date before you pay anything.