Model

HeyGen sells Avatar 6 while Tavus's Griffin scores 3.83 against a human's 3.92 and still has no price

HeyGen put Avatar 6 on general availability on November 17 with a public API and first place in New Measure's blind Avatar Arena; Tavus's Griffin, which Tavus says is the highest score ever recorded for an AI on NVIDIA's face-to-face benchmark, is still a preview with no price. Revenue accrues to HeyGen today because Avatar 6 is the model on sale; Tavus's best model is not.

Vincent JiangVincent Jiangthree-minute read
Share
Joshua Xu, co-founder and chief executive of HeyGen, speaking on stage at Collision 2024 in Toronto
Joshua Xu, HeyGen's co-founder and chief executive, on the PandaConf stage at Collision 2024 in Toronto. HeyGen put its Avatar 6 model on general availability on November 17, 2026.

Seven weeks after Tavus, a San Francisco AI research lab, unveiled Griffin on October 1 as the first "Human Interaction Model," a full-duplex model that holds live face-to-face conversations, its own site still lists "Griffin (Preview)" in the model menu, with no price and no release date 12. Only Griffin-Lite has shipped, to select trusted testers, while Tavus builds the disclosure and safety mechanisms it says the model needs 1.

Tavus has raised $70M from CRV, Sequoia Capital, Scale Venture Partners and Y Combinator; its 150,000-plus developers and customers include Amazon−2.75% — Amazon, down 2.75 percent today, Mayo Clinic and Salesforce−0.18% — Salesforce, down 0.18 percent today 1. What they can buy today is a generation older: self-serve plans from $22 to $975 a month run the Phoenix-4.5 pipeline, and Griffin appears in no tier 3.

The scores lead the field, the sales do not

On NVIDIA's VideoFDB benchmark of face-to-face AI, Griffin scored 3.83 out of 5 on generation, 0.09 below the 3.92 human reference; the next-highest published system managed 2.80 1.

Griffin scores 3.83 on NVIDIA's VideoFDB, 0.09 below the human reference

The next best published system is more than a point behind.

0/51/52/53/54/5Human reference3.92/5Tavus Griffin3.83/5Next best published AI2.8/5
Data
Value
Human reference3.92/5
Tavus Griffin3.83/5
Next best published AI2.8/5
Generation scores on NVIDIA's VideoFDB benchmark of face-to-face AI, 5-point scale, as reported in Tavus's October 1, 2026, announcement.1

The 48% headline comes from a live study Tavus ran itself: of 54 participants who held one-minute video calls with Griffin, 48% believed they had been talking to a human 14. The previous Tavus stack convinced one rater in 41, or 2.4% 4.

HeyGen's Avatar 6 is already on sale

HeyGen~$500M — HeyGen, private, latest valuation ~$500M, a rival whose platform turns scripts into presenter videos, put Avatar 6 on general availability on November 17 with a public API 56, and says it doubled to $200 million in annual recurring revenue in the eight months to June 5. It tops New Measure's blind Avatar Arena, a chess-style preference rating: 1,197 Elo on personal avatars, 1,237 on stock presenters 6.

Every headline number carries an asterisk

The 48% has no independent confirmation 7. Maty Bohacek, a Stanford researcher, told the publication AI Engineer that 99% of participants caught Tavus's previous model in his academic test, and that AI agents impersonating people beat phone and Zoom−2.18% — Zoom, down 2.18 percent today verification in 65% of runs 7.

HeyGen's banner 84.4% is arithmetic: raters picked Avatar 6 as the real recording in 42.2% of side-by-side viewings, and the advertised figure doubles that 6. The arena's clips ran 35 seconds or less in 97% of comparisons, on renders that typically took about three minutes 6.

The next number that matters is a price

Tavus is monetizing a generation its own benchmark numbers say it has surpassed, while Avatar 6 is the model a buyer can put on a credit card 36. The capital behind both is venture money: Tavus's $70M from CRV, Sequoia, Scale and Y Combinator 1, and HeyGen's $60M round at a $500 million valuation led by Benchmark in June 2024 8.

The signal to watch is pricing. When Tavus publishes a Griffin rate card, buyers can compare a live-conversation model against Avatar 6 on cost, not on claims 6.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading AMZN

Reader comments

0 comments

    Model

    Pika's Enact lets creators direct AI video mid-scene, with answers in under a second

    Enact, posted November 18, lets a creator drag an AI actor mid-scene, talk to the cast by name and redo the camera after locking a performance. Pika open-sourced the model underneath and kept the directing layer closed, a bet that value in AI video is moving from model quality to workflow.

    Model

    Anthropic falls 11% toward its $83 IPO price as Microsoft undercuts Claude

    Anthropic fell about 11% Friday morning, back toward the $83 IPO price, as Microsoft's price attack on Claude met the open market. Microsoft holds Anthropic's compute contract, its shares and a cheaper rival product; a leveraged income ETF on the two-day-old stock listed the same morning.

    Model

    Ramp's invoices show Anthropic held 58.7% of US business AI spend vs OpenAI's 37.2% before its record IPO

    Ramp's index of 85,000 US business bill-payers is the first independent check on the record $81.3 billion IPO: the enterprise lead over OpenAI is real, and so is the falling price of what Anthropic sells.

    Model

    Google's Gemini 4 Neon took the coding lead the day before Anthropic's $1.5 trillion debut

    Google's mid-tier model debuted at No. 1 on Artificial Analysis's coding-agent index at $6.86 a task against Claude Opus 5.5's $12.61, a day before Anthropic's Nasdaq debut. A $1.5 trillion valuation built on coding-led growth now meets a measured, cheaper rival, and the exposure sits with whoever paid $83.

    Model

    ElevenLabs, valued at $22 billion, puts its voice agents inside NICE, Genesys, Amazon Connect and Five9

    Salesforce and Five9 have put more than $3 billion into buying customer-service AI this month. ElevenLabs, valued at $22 billion after a September share sale, embedded its agents natively into Five9's platform and three rivals instead, and the evidence points to pricing power sitting with the channel, not the model maker.

    Model

    OpenAI's agent bill now runs at twice its researchers' cash pay

    OpenAI's own report prices agent use at twice its researchers' cash pay while internal deployment is 2 percent of research compute, and the number lands the week Anthropic lists and OpenAI settles for half its raise. The bill stacks inference on the chip budget, and holders pay it.

    Subscribe to TheInference for free