Google Gemini 3.5 Transcribe lands #5 on AA-WER with 2.6% error rate

As seen on the 24/7 Wall St. homepage on August 26, 2026.

Artificial Analysis @artificialanlys

Google is now pricing into the same speech-API lane OpenAI, ElevenLabs and AWS are fighting over, and a 0.40s streaming lag is the number that wins voice-agent contracts.

Google has released Gemini 3.5 Transcribe, ranking #5 on AA-WER at 2.6%, alongside Gemini 3.5 Transcribe Live, achieving 4.0% AA-WER Streaming at 0.40s after speech end @GoogleAI's Gemini 3.5 Transcribe release comprises two API offerings: Gemini 3.5 Transcribe Live for https://t.co/xOtoqMOY1s
  • Replies4
  • Reposts3
  • Likes30
Continue ReadingShow less

Google launched two new speech API offerings on August 26: Gemini 3.5 Transcribe, which scored a 2.6% word error rate on the Artificial Analysis benchmark to claim the fifth spot on the AA-WER leaderboard, and Gemini 3.5 Transcribe Live, designed for real-time streaming use cases.

Gemini 3.5 Transcribe Live posts a 4.0% AA-WER Streaming score with a latency of just 0.40 seconds after speech ends, the kind of number that determines whether a voice-agent product feels responsive or sluggish in production.

Sponsored

_________________________________

What's Your Number...?

Here's a question most people 5y from retirement can't answer: at your current savings rate, how much do you need, and how long will it actually last? A good advisor can put a date on that in a single meeting. SmartAsset's free quiz matches you with up to three fiduciary advisors serving your area, so you can get YOUR retirement number now (sponsor)

__________________________________________

Both models are available through Google's API, meaning developers building voice assistants, transcription pipelines, or real-time call analytics can access them today. Google is stepping directly into a market where OpenAI, ElevenLabs, and AWS are already competing for the same speech-API contracts.

In voice-agent deployments, latency of 0.40 seconds decides enterprise deals, and it sets a concrete benchmark that rivals will now have to answer.

Sources

Related Market Updates

Jensen Huang
Earnings · 2027 Q2 Jensen Huang CEO, NVDA
AI has reached its inflection point. It's doing useful work. Its tokens are productive and profitable. Now, compute is revenue.

Huang is reframing the AI trade on the fiscal Q2 call: customer compute is now billable output, the argument that keeps a $5.1 trillion market cap intact. Anyone underwriting a demand slowdown has to answer that claim first.

Read more
Jensen Huang
Earnings · 2027 Q2 Jensen Huang CEO, NVDA
And demand is accelerating. This time last year, one lab alone was driving the buildout; today, we have a golden age of new AI labs and startups, multiple frontier labs scaling in parallel, a thriving open-model ecosystem and physical AI coming online — with strong momentum across the U.S. and around the world. The AI infrastructure buildout is at full steam. Vera Rubin, now in full production, was built to power exactly this moment.

Huang's fiscal Q2 message shifts the demand story from a single anchor customer a year ago to frontier labs, startups and open models buying in parallel. Vera Rubin reaching full production means the next platform cycle is already shipping into that base.

Read more
Artificial Analysis @artificialanlys

Alphabet is now selling speech-to-text at 0.40s latency after speech ends, pushing Gemini straight into the voice API market that Whisper and Deepgram have owned.

Google has released Gemini 3.5 Transcribe, ranking #5 on AA-WER at 2.6%, alongside Gemini 3.5 Transcribe Live, achieving 4.0% AA-WER Streaming at 0.40s after speech end Google’s Gemini 3.5 Transcribe release comprises two API offerings: Gemini 3.5 Transcribe Live for continuous https://t.co/y81oauk7Sj
  • Replies1
  • Reposts0
  • Likes0
Read more
Artificial Analysis @artificialanlys

Nine cents a task with a frontier-adjacent score is the number squeezing US model pricing, and it lands while OpenAI and Anthropic are already cutting to hold ground.

GLM-5.3-Flash scores 57 on the Artificial Analysis Intelligence Index. At $0.09 Cost per Task, it sits comfortably on the Intelligence vs. Cost per Task Pareto frontier @Zai_org has released GLM-5.3-Flash, a smaller and cheaper sibling to GLM-5.3 at 320B total parameters and https://t.co/YJRMoU02oK
  • Replies3
  • Reposts9
  • Likes76
Read more
Jim Cramer @jimcramer

Cramer's math: a $12.6 billion resolution against a feared $200 billion, with no operating restrictions attached, removes the overhang that had been priced into Meta.

Meta was looking at a trillion; betting like was $200 billion and big restrictions: final is $12.6b and no onerous restrictions
  • Replies15
  • Reposts0
  • Likes24
Read more
Chamath Palihapitiya @chamath Quote post

If enterprises keep routing work to cheaper models, the frontier labs burning billions on training have a payback problem, and Chamath is pointing at the data showing Anthropic's top model draws just 11% of revenue.

Interesting development. The latest models are also the most expensive. One can see a long J curve building, then, to get to positive ROI for these models if folks don’t want to pay for the latest/greatest. On top of that, if most enterprises are realizing that previous https://t.co/ANKAC2zGm1 [Quoted @kyleichan]: This is the most important chart in AI right now because it shows where AI spending is actually going by model capability. Of Anthropic’s models, Fable (dark blue bottom) is only 11% of revenue. Very few companies want to pay for the best model. The majority of spending is https://t.co/oCIehWk0aS
  • Replies5
  • Reposts3
  • Likes24
Read more
Jim Cramer @jimcramer

Cramer's frame ahead of tonight's earnings report: Nvidia trades at just 16 times FY2028 earnings once you account for the January fiscal year end.

Remember NVDA fy ends Jan31, so Jan 31 2027 IS 2027, important because Jan FY 2028 16x pe. this is second q of FY 2027 reported tonight. SIXTEEN TIMES FULL YEAR FY 2028. Next four quarters 18.7
  • Replies40
  • Reposts12
  • Likes193
Read more
Jim Cramer @jimcramer

Cramer stops just short of an endorsement on Vertiv, crediting its closed-loop cooling as an overlooked edge while data-center water use draws scrutiny from Texas to Austin.

Vertiv has a solution for water use that makes it circular. But nobody talks about it. Electricity still high but that can be worked out with communities by paying their electric bills. Would be pounding table on Vertiv but political issues....
  • Replies22
  • Reposts4
  • Likes142
Read more
Jim Cramer @jimcramer

Cramer's take lands as MLPerf benchmark headlines pit AMD's Instinct line and custom silicon against Blackwell, and Nvidia's next earnings report is the scoreboard that settles it.

Every day i read about some chip that is superior to Nvidia. And every year i see no real competitors
  • Replies52
  • Reposts5
  • Likes220
Read more
Elon Musk @elonmusk Repost

A single booster flying 37 times pushes SpaceX's cost per launch further below anything rivals can match, and every legacy launch provider and space ETF holding is priced against that gap.

RT @SpaceX: Falcon 9 lands on the A Shortfall of Gravitas droneship, completing the first 37th launch and landing of a booster https://t.co…
  • Replies0
  • Reposts598
  • Likes0
Read more
Elon Musk @elonmusk Repost

A third Starship launch complex in Louisiana adds to the Texas and Florida sites, and every rival spaceport operator now has to price in a competitor building cadence at industrial scale.

RT @SpaceX: Introducing Starbase, Louisiana – a high-cadence launch site that will enable Starship to open a pathway to the stars for human…
  • Replies0
  • Reposts1.3K
  • Likes0
Read more