A homegrown text-to-speech model just posted naturalness scores above OpenAI’s Realtime API and ElevenLabs’ fastest voices — while billing at a fifth of the price. Here’s what’s verified, what to watch, and what it means for Indian BFSI and BPOs.
Murf AI Falcon 2 went live to the public today, and the numbers attached to the launch are the kind that usually belong to a company ten times its size. The Bengaluru-built text-to-speech model has posted naturalness scores above OpenAI’s Realtime API and two of ElevenLabs’ most widely used voices on Artificial Analysis — the independent benchmark the voice-AI industry treats as its scoreboard — while charging roughly a fifth of what those rivals bill per minute of generated speech.
The claim comes from Bloomberg’s reporting on the launch, published the night before Falcon 2’s public rollout, and it’s a rare instance of an Indian AI company being benchmarked directly against Silicon Valley’s best-funded voice labs — and coming out ahead on at least one axis that actually matters to a buyer: naturalness per dollar.
What Is Murf AI Falcon 2, Exactly?
Falcon 2 is a streaming text-to-speech (TTS) foundation model — not a chatbot, not a language model that reasons, just a system built to turn text into speech instantly and convincingly enough that a caller can’t tell it’s synthetic. It’s the second generation of Murf’s Falcon line, which first launched in November 2025 with a claimed 55-millisecond model latency. Falcon 2 pushes that further: Murf says it now returns first audio in under 100 milliseconds, fast enough that a phone-based conversation doesn’t feel like it’s talking to a machine on a satellite delay.
On paper, the spec sheet reads like this:
- 150+ voices across 35+ languages
- Instant voice cloning from as little as 10 seconds of reference audio
- 10,000+ concurrent calls supported without degrading audio quality
- Data residency across 11 geographies, plus an on-premise deployment option for regulated industries
- Already used in production by Honeywell, Cisco, Pfizer, VMware, Nestlé and Air France-KLM, according to Bloomberg’s reporting on Murf’s client list
Falcon 2 was in limited release to select enterprise clients before today’s broader public availability, which is standard practice for a model this latency-sensitive — you want it battle-tested on real call volume before opening the gates.
The Benchmark Claim, Checked
Here’s the part worth being precise about, because “beats OpenAI and ElevenLabs” is doing a lot of work in every headline about this launch, including ours.
Murf isn’t claiming the #1 spot on the overall Artificial Analysis Speech Arena leaderboard — that board is currently a tight, weekly-shuffling contest between models like Cartesia’s Sonic series and Speechify’s Simba, among others. What Murf is claiming, and what Bloomberg’s independent reporting corroborates, is narrower and still genuinely impressive: on Artificial Analysis’s naturalness scoring, Falcon 2 ranks above ElevenLabs’ Flash v2.5 and Turbo v2.5 voices, OpenAI’s Realtime API, and xAI’s TTS offering.
That’s a real result on a real, independent, listener-voted benchmark — not a number Murf made up internally. But two caveats belong in the same sentence as the claim:
- Leaderboard rankings on Artificial Analysis move frequently as labs ship updates; “ranks above X and Y today” is a snapshot, not a permanent title.
- Naturalness is one axis. It doesn’t automatically tell you how a model performs on your specific accent, your specific script full of policy numbers and IFSC codes, or your specific call-center noise floor.
Murf AI Falcon 2 Pricing — How the Math Actually Works
The number that will actually move enterprise buying decisions is cost, and here Falcon 2’s gap over the field is large enough to matter at scale.
| Model | Price / minute | Price / 1M minutes | Time to first audio |
|---|---|---|---|
| Murf Falcon 2 | $0.01 | $10,000 | <100 ms |
| ElevenLabs Turbo / Flash | ~$0.05 | $50,000 | 150–300 ms (typical) |
Per-minute figures for Falcon 2 and the $10,000 / $50,000 per-million-minute comparison are as reported by Bloomberg. ElevenLabs’ own published API rate is $0.05 per 1,000 characters for Flash/Turbo (elevenlabs.io/pricing/api) — which, at a typical speaking pace, converts to roughly the same $0.05/minute figure Bloomberg cites. ElevenLabs’ time-to-first-audio is not independently benchmarked in this piece; verify against current documentation before quoting in a procurement deck.
*Illustrative conversion at ₹95.6/$1 (Aug 20, 2026 mid-market rate). A gap of roughly ₹38 lakh per million minutes generated — the kind of volume a mid-sized outbound-calling or IVR operation could plausibly hit in a few months, not a hypothetical enterprise-only number.
Who’s Behind Murf AI
Murf AI was founded in October 2020 by three IIT Kharagpur classmates — Sneha Roy, Ankur Edkie and Divyanshu Pandey — starting out as a voiceover tool for presentations and video before pivoting toward the real-time, developer-facing API business that Falcon and now Falcon 2 represent. The company has raised a modest $11.5 million total: a $1.5 million seed round led by Elevation Capital in 2021, followed by a $10 million Series A led by Matrix Partners India in 2022, with Elevation returning to participate. That’s a small war chest next to ElevenLabs, which has raised well over $100 million, or OpenAI’s balance sheet — which is exactly why the benchmark result is getting attention.
“For years, the most important AI models have largely emerged from the US.” — Sneha Roy, COO, Murf AI, in a statement to Bloomberg
Roy’s framing of Falcon 2 is that it’s engineered specifically around the economics of enterprise production use — not a research demo optimized to win a leaderboard, but a system built to be cheap enough and fast enough to run at call-center scale.
The India Angle: Why This Matters Beyond a Benchmark Screenshot
This is where the story stops being a Silicon Valley leaderboard update and starts being relevant to an Indian founder’s or CTO’s actual roadmap.
The BPO and BFSI math
India’s call-center and BPO industry is precisely the buyer Falcon 2 is built for: high call volume, thin margins, and a genuine appetite to automate reminder calls, collections, KYC verification prompts and basic customer queries without sounding like an IVR system from 2012. At $10,000 per million minutes versus $50,000, the price gap alone could be the difference between a voice-AI pilot that gets funded and one that gets shelved after the first cost review.
Built with India’s developer community, not just for it
Murf ran a public “10 Days of Voice Agents — #VoiceForBharat” builder challenge between August 6 and 15, tasking developers with building a voice agent capable of handling a real phone call for someone in India by Independence Day — using Falcon 2 as the underlying engine. It’s a small signal, but it shows Murf positioning Falcon 2 as infrastructure for the Indian voice-agent developer ecosystem, not just an enterprise sales pitch aimed at Fortune 500 logos.
Where it sits next to India’s other voice-AI bets
Falcon 2 isn’t launching into empty space. Sarvam AI’s Bulbul model and the broader IndiaAI Mission-backed sovereign-AI push are also targeting Indian-language voice interfaces, and global players like Cartesia have been adding Indian-language support to their own leaderboard-topping models. Falcon 2’s pitch is less about being India-specific in language coverage and more about being India-priced at global quality — a different, and arguably more immediately commercial, bet.
Reality Check: What to Verify Before You Switch
Test on your own data. Benchmark naturalness scores are measured on standardized prompts — not on your specific accent mix, your compliance-heavy scripts, or your background-noise conditions. Run a pilot before committing.
Leaderboard rank is a snapshot. Artificial Analysis rankings shift as competitors ship updates; the “beats OpenAI and ElevenLabs” claim reflects today’s board, not a permanent guarantee.
Cost comparisons mix units. Falcon 2’s pricing is quoted per minute; ElevenLabs’ official rate is quoted per character. The conversion above is a reasonable estimate, not a like-for-like guarantee — get an actual quote for your usage pattern before budgeting.
Indian-language depth is unverified here. Murf claims 35+ languages, but this piece has not independently tested Falcon 2’s Hindi, Tamil or other Indic-language output quality against Sarvam or Cartesia’s Indian-language voices. Treat that as an open question, not a settled one.
What to Watch Next
Three things worth tracking over the coming weeks: whether Falcon 2 climbs onto the public Artificial Analysis leaderboard in its own right rather than being cited via company-supplied comparisons; whether Murf discloses any enterprise contract wins specifically tied to Falcon 2 rather than the broader Murf platform; and whether rivals respond with price cuts of their own, which is the usual next move whenever a credible new entrant undercuts the market by 4-5x.
FAQ: Murf AI Falcon 2
What is Murf AI Falcon 2?
Falcon 2 is a real-time text-to-speech foundation model built by Bengaluru-based Murf AI, designed for enterprise voice agents, call centers and IVR systems. It became publicly available on August 20, 2026.
How much does Murf Falcon 2 cost?
Falcon 2 is priced at $0.01 per generated minute, or $10,000 per million minutes — roughly a fifth of ElevenLabs’ Turbo/Flash rate of about $0.05 per minute, per Bloomberg’s reporting.
Is Murf Falcon 2 better than ElevenLabs and OpenAI?
On the independent Artificial Analysis Speech Arena’s naturalness scoring, Falcon 2 ranks above ElevenLabs’ Flash v2.5 and Turbo v2.5 voices and OpenAI’s Realtime API. It is not the #1 model on the overall leaderboard, which remains a close contest among several providers.
Who founded Murf AI?
Murf AI was founded in October 2020 by IIT Kharagpur alumni Sneha Roy, Ankur Edkie and Divyanshu Pandey. The company has raised $11.5 million total from Matrix Partners India and Elevation Capital.
Does Falcon 2 support Indian languages?
Murf says Falcon 2 supports 150+ voices across 35+ languages. This piece has not independently verified the depth of its Hindi or other Indic-language quality against India-focused rivals like Sarvam AI.