Menlo’s keyboard-kill bet hits $2 billion with Wispr

Wispr raised $280 million in Series B funding at a $2 billion valuation on August 17, led by Menlo Ventures, as its AI dictation app Flow passed 60 billion words written by users. The round lifts total capital to $361 million and arrives with a preview of Canto, the company’s first proprietary speech model.

Menlo partner Matt Kraning framed the check as more than a follow-on. Fourteen months after arguing Wispr would kill the keyboard, the firm is now betting the text box is the next interface to disappear.

Menlo doubles down after 30x revenue growth

Kraning wrote that Wispr has grown revenue over 30x year over year since the earlier thesis piece. Flow now runs in 162 countries and more than 100 languages across tens of thousands of paying businesses. He called the Series B one of the largest investments Menlo has made in an AI company.

Existing backers Notable Capital, NEA, Neo Ventures, 8VC and MVP Ventures returned. New money came from Acrew, Activate, Forerunner, Goodwater, Peak XV, Together Fund, PLUS Capital and a group of athletes and cultural figures that includes Livvy Dunne, Shaun White, Dak Prescott, DK Metcalf, Joe Burrow, Klay Thompson, Paul George and Trae Young.

• $280 million Series B at $2 billion post-money
• $361 million total raised to date
• 60 billion+ words written with Flow
• Users at nearly all Fortune 500 firms and over 10,000 enterprises

Founders Tanay Kothari and Sahaj Garg, Stanford batchmates who started the company in 2021 after multiple pivots, said the money first goes to accuracy research. “The whole reason to talk instead of type is to stay inside your own train of thought,” Kothari wrote in the company post announcing $280 million in Series B funding.

That framing turns the raise into a product mandate rather than a balance-sheet event. Capital follows the claim that dictation only wins when the user never leaves the thought mid-sentence to fix a transcript.

How the valuation climbed from $700 million

The path was short and steep. Menlo led a $30 million Series A in mid-2025. In November 2025 Notable Capital led a $25 million extension that valued Wispr at roughly $700 million and brought cumulative funding to $81 million. Less than ten months later the company is nearly triple that mark.

Kothari noted that a year earlier most conversations about voice still started with skepticism built on 15 years of disappointing dictation tools. Those talks have largely stopped. Users now ask what Flow should do next.

The compression of that arc matters for how the new round will be read. Menlo led early, stepped back for the extension, then returned as lead at a multiple that prices Flow as infrastructure rather than a niche app. Returning funds and a long list of new names both signal that the growth story survived a full product cycle between checks.

Canto cuts real-world error rates fourfold

Alongside the raise Wispr previewed Canto, a 2-billion-parameter speech model built for noisy cars, open offices, wind and heavy accents rather than clean studio audio. In the hardest conditions word error rates fall from more than 30 percent to between 5 and 10 percent. Across everyday use the company expects 30 to 35 percent fewer dictations that need any editing.

The model also handles code-switching. Roughly half the world mixes languages inside a single day or sentence. Hinglish is the clearest example: Hindi speakers usually write Devanagari but expect romanized output when speaking the hybrid, so correct recognition alone is not enough. Canto draws on personal dictionaries and names around the user.

I use Flow every day. I grew up rotating between English, Spanish, and Lithuanian, and it keeps up with me no matter which one I’m speaking. So the opportunity to invest and support a product I use and believe in was a no brainer.

Domantas Sabonis, three-time NBA All-Star and investor, said that after testing the product himself.

Internally Wispr tracks “zero edit rate,” the share of utterances that return completely untouched and ready to send. Most of the new capital targets closing that gap.

Accuracy research therefore sits upstream of every other bet in the round. A model that fails in a car or an open office forces the user back to the keyboard, which undoes the train-of-thought pitch Kothari put at the center of the announcement.

India became the second-largest market

India is now Wispr Flow’s second-biggest market by users and revenue after the United States. Growth there hit roughly 100 percent month-over-month after the Hinglish rollout and a local campaign, up from about 60 percent earlier. The company has added local hires and treats code-switching as a core requirement rather than an edge case.

Enterprise pull has been organic. Teams deploy Flow without a heavy sales motion, then expand it company-wide. Android launched earlier in 2026; Windows support for the new meeting notetaker is coming. The notetaker, released in early August, sits alongside Flow and competes with Granola, Otter and Fireflies by turning spoken meetings into summaries and action items.

The India numbers also test whether Canto’s design choices travel. A market that mixes scripts and languages inside a single utterance is a harsher proving ground than clean English office audio. Local hires and a dedicated campaign show the company is treating that market as a product surface, not only a growth chart.

The lab that aims past the text box

In July Wispr opened the Wispr Advanced Interfaces Lab under chief scientist Ariya Rastrow, a founding member of the original Amazon Alexa team who later worked on multimodal models at Meta. The lab’s brief is voice-to-outcome systems that understand intent in full digital context and respond with actions, workflows or generated interfaces instead of another block of text.

Kraning argued the bottleneck in AI has moved from the model to the interface. Frontier labs deliver superhuman intelligence, yet most people still talk to it through a text box that taxes thought. Flow is the first wedge; the lab is meant to turn that wedge into a lasting channel for digital thought. Similar pressure is visible in full-duplex voice interfaces that feel human elsewhere in the industry.

Engineers already steer coding agents by voice because typing prompts to a system that types faster is the new bottleneck. Sales teams talk through a call and watch the CRM update. Those pulls are what convinced Menlo the original keyboard thesis was incomplete.

Rastrow’s path from Alexa to multimodal work at Meta maps onto the lab’s charter. The group is not chasing cleaner transcripts alone. It is trying to collapse the step between a spoken request and a finished action inside the tools people already open every day.

Competition arrives from every direction

Google launched Rambler, a Gemini-powered dictation feature inside Gboard, earlier in 2026. It cleans filler words, handles multilingual switches and accepts voice edit commands. Free OS-level tools and lower-priced apps such as Willow, Superwhisper, Monologue, Aqua and Typeless crowd the same shelf. Some users had already flagged a temporary quality dip in Flow before Canto’s preview.

Menlo’s counter is that “free and mostly right” loses to “zero edits, everywhere” for anyone who talks for a living, and that acting on intent requires context no OS checkbox ships. Distribution and product maturity still separate the leaders even when base models converge. The same dynamic shows up among other AI startups racing to multi-billion tags that pair proprietary data loops with clear workflow products.

• OS-level free tools (Rambler inside Gboard)
• Lower-priced specialist apps (Willow, Superwhisper, Monologue, Aqua, Typeless)
• Meeting notetakers (Granola, Otter, Fireflies) on the adjacent surface

The temporary quality dip some users reported before the Canto preview is a reminder that dictation products live or die on daily trust. A free feature that is usually good enough can still win casual use. Paid power users, Menlo argues, will keep paying when the alternative is another round of edits.

What the next capital is actually buying

Kothari closed the announcement by saying the years ahead will be judged on how much of the distance between spoken word and perfect output the company closes. Most of the $280 million funds the research that moves the zero-edit rate and the work of putting that model everywhere people already talk and type.

The keyboard bet, in Menlo’s telling, paid off on schedule. The larger wager is now live: that the text box is the next layer to vanish once systems stop returning paragraphs and start delivering outcomes. Wispr has the capital, the model preview, the lab and the usage numbers. The scoreboard remains the share of dictations that never need a single keystroke.

Flow Moves Onto Desktops and Into Meetings

Dictation alone no longer defines the product surface. Android arrived earlier in 2026, widening the path beyond the first wave of users. Windows support for the meeting notetaker is next, which pulls Flow into the machines where longer work still happens.

The notetaker, shipped in early August, turns spoken meetings into summaries and action items. That puts Wispr next to Granola, Otter and Fireflies on a second shelf while Flow keeps owning the blank-page moment. One product captures thought as it forms; the other captures decisions as a room talks.

1. 2021 – Kothari and Garg found Wispr after multiple pivots
2. June 2025 – Menlo leads $30 million Series A
3. November 2025 – Notable Capital leads $25 million extension near $700 million
4. Earlier 2026 – Android launches
5. July 2026 – Wispr Advanced Interfaces Lab opens under Ariya Rastrow
6. Early August 2026 – Meeting notetaker releases
7. August 17, 2026 – $280 million Series B at $2 billion; Canto previewed

Each step widens where a spoken sentence can land. Phones, desktops, live meetings and agent workflows all become targets for the same accuracy stack the Series B is meant to fund.

Organic Enterprise Adoption Keeps Compounding

Enterprise demand has not required a heavy sales motion. Teams install Flow, prove it on real work, then spread it company-wide. That pattern now reaches users at nearly all Fortune 500 firms and more than 10,000 enterprises, alongside tens of thousands of paying businesses in 162 countries.

Organic expansion changes what the new capital must do. Less of the round has to buy pipeline; more of it can fund the accuracy research and platform work that keep those quiet rollouts from stalling. When a team already trusts the product, the next feature has a ready path in.

Sales crews who talk through a call and watch the CRM update, and engineers who steer coding agents by voice, are concrete versions of the same loop. Spoken input becomes a finished artifact inside software the company already owns. That is the habit Menlo is underwriting at $2 billion.

Frequently Asked Questions

What is Wispr Flow’s Canto speech model?

Canto is Wispr’s first proprietary 2-billion-parameter speech model, previewed with the Series B. It is trained for real-world noise, accents and code-switching rather than clean studio audio, cutting hardest-condition word error rates from over 30 percent to 5-10 percent and expected to cut everyday edits by 30-35 percent.

How much total funding has Wispr raised?

The $280 million Series B brings Wispr’s cumulative capital to $361 million. Earlier rounds included a $30 million Series A led by Menlo in 2025 and a $25 million extension led by Notable Capital that valued the company near $700 million.

Who founded Wispr and when?

Tanay Kothari and Sahaj Garg, Stanford classmates, founded the company in 2021. After several pivots they landed on the Flow dictation product that became the core business.

What does Wispr mean by zero edit rate?

Zero edit rate is the internal metric for the share of spoken input that returns completely correct and ready to send with no user correction at all. It is stricter than “close enough” or “quick to fix” and is the standard the company says it will be judged against.

Why is Hinglish support important for Wispr?

India is Wispr’s second-largest market. Hindi speakers typically write in Devanagari but expect romanized output when speaking Hinglish. Models that only recognize words fail to produce sendable text; Canto and the localization work address that gap and helped drive 100 percent month-over-month growth after launch.