Industries / Consumer goods / Eyewear
Smart glasses
Glasses that carry cameras, microphones, speakers, displays or an AI assistant, made to be worn like ordinary spectacles or sunglasses, often with prescription lenses.
7
Milestones achieved, from 1980s to 2025.
3
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1980s
Computers worn on the head
Steve Mann wore homemade head-mounted cameras, displays and computers in daily life in the 1980s, when nobody else did. IEEE Spectrum calls him the father of wearable computing. Accounts differ on where the early work happened: IEEE Spectrum places it in high school and at McMaster University, while the University of Toronto says he began prototypes in the 1980s as a graduate student at MIT.
Credit
Steve Mann
2013
Google Glass on sale
Google began shipping Glass Explorer Edition, glasses with a camera and a small display, in April 2013 at US$1,500 each, starting with about 2,000 people who had pre-ordered at its developer conference. It stopped selling Glass to individuals after 19 January 2015, after privacy concerns and without a convincing everyday use, and ended sales of Glass Enterprise Edition, a version for workplaces, on 15 March 2023.
LessonA camera on the face needs acceptance from the people nearby and a clear job to do. Glass was barred from some bars, and a workplace version alone did not keep it going.
Credit
Google
2016
Spectacles and a write-down
Snap sold its first Spectacles, sunglasses with a camera for short videos, from vending machines called Snapbots for US$129, starting in November 2016. In the third quarter of 2017 it recorded US$39.9 million of charges for Spectacles inventory, mainly for excess stock and cancelled purchase commitments.
LessonExcitement built on scarcity did not become lasting demand, and the maker carried the cost of unsold stock.
Credit
Snap Inc
2020
Glasses left unsupported
Google bought North, the Canadian maker of Focals display glasses, in June 2020. North cancelled Focals 2.0, refunded Focals buyers, and ended software support for Focals 1.0 on 31 July 2020. Bose began phasing out its Frames audio glasses in 2023.
LessonBuyers of glasses from a small maker can be left with a device nobody supports when the maker is bought or leaves the field.
Credit
North; Google; Bose
2023
Camera glasses sell widely
Meta and EssilorLuxottica released Ray-Ban Stories at US$299 in September 2021, then Ray-Ban Meta glasses with a 12 MP camera and the Meta AI assistant, on sale from 17 October 2023. Both could be fitted with prescription lenses. The research firm Counterpoint reported that global smart glasses shipments grew 110 percent year on year in the first half of 2025 and 139 percent in the second half, when Meta's share reached 82 percent. Xiaomi, whose AI glasses went on sale in China in June 2025 at ¥1,999, ranked second.
about.fb.comabout.fb.comcounterpointresearch.comcounterpointresearch.comroadtovr.com
Credit
Meta; EssilorLuxottica
2024
The recording light falls short
In September 2021 the Irish Data Protection Commission and Italy's data protection authority questioned whether the "very small" light that shows Ray-Ban Stories is recording did its job, noting that comprehensive field testing had not been shown to them. In October 2024 two Harvard students, AnhPhu Nguyen and Caine Ardayfio, showed that Ray-Ban Meta glasses linked to the PimEyes face search could go from a stranger's face to a name. In September 2026 a report by the Hamburg data protection authority found the light "not easily visible in many environments, particularly outdoors and in direct sunlight".
LessonA warning light the maker designed is not proof that the people nearby notice it; that has to be tested where the glasses are worn.
Credit
the Irish Data Protection Commission and Italy's Garante; AnhPhu Nguyen and Caine Ardayfio; the Hamburg Commissioner for Data Protection
2025
A display and a wristband
Meta put Meta Ray-Ban Display on sale in the US on 30 September 2025: glasses with a small display in the lens, controlled by the Meta Neural Band, a wristband that reads the electrical signals of the wrist muscles (EMG). The pair cost US$799. Prescriptions from −4 to +4 dioptres are supported, UploadVR reported. Engadget's review found the band almost never missed a gesture but the frames thick. In January 2026 Meta paused launches in the UK, France, Italy and Canada, citing demand and limited stock.
Credit
Meta; EssilorLuxottica
Watch for
Android glasses from eyewear brands
Google and Samsung opened pre-orders on 12 November 2026 for Android XR glasses with a 12 MP camera and a recording light, in frames by Warby Parker (from US$395) and Gentle Monster (US$449 to US$489), with prescription options and availability from 10 December.
Achieved whenunits reach customers and independent reviews are published.
Claimed by
Google and Samsung, with Warby Parker and Gentle Monster, November 2026
Snap's Specs on sale
Snap says its consumer Specs, augmented-reality glasses that lay images over the wearer's view, cost US$2,195 and take removable prescription inserts, with shipping in the US, the UK and France expected from October 2026. No report of units reaching customers was found.
Achieved whencustomers receive units and independent reviews appear.
Claimed by
Snap, June and September 2026
Ten million pairs a year
EssilorLuxottica's chief executive Francesco Milleri said in February 2025 that the company would be able to make 10 million smart glasses a year by the end of 2026. Road to VR, citing Bloomberg, reported in January 2026 that Meta had proposed 20 million, and possibly more than 30 million. EssilorLuxottica says it sold more than 7 million AI glasses in 2025.
Achieved whenan analyst publishes unit volumes that match, or an audited filing states that the capacity was reached.
Claimed by
EssilorLuxottica, February 2025 and February 2026; Meta, January 2026, as reported
Open
01
Caption glasses, scored
A public, repeatable test of live-caption glasses for deaf and hard-of-hearing users. It measures word error rate, the share of words captioned wrongly, and caption delay in quiet, café and street noise, and collects ratings from deaf and hard-of-hearing testers.
The briefCredit · unclaimed02
Seeing the recording light
A controlled test of whether people notice the recording light on camera glasses at 1, 2, 4 and 8 metres and several angles, indoors and outdoors in daylight, with the light level measured, and a published tally.
The briefCredit · unclaimed03
Describing scenes for blind users
A test of AI scene description on camera glasses with blind and low-vision participants, on a fixed set of 200 real scenes such as signs, labels, door numbers and transit boards, scored by category. Money-handling tasks are left out.
The briefCredit · unclaimed
Open 01 · the brief
Caption glasses, scored
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, repeatable test of live-caption glasses for deaf and hard-of-hearing users. It measures word error rate, the share of words captioned wrongly, and caption delay in quiet, café and street noise, and collects ratings from deaf and hard-of-hearing testers.
Why it is still open
The accuracy figures published for caption glasses are the makers' own: XanderGlasses "85% to 95%" depending on noise, LEION "up to 98%" and AirCaps "97%", as collected by HearingTracker, which reports no controlled test. XanderGlasses' own help page gives no test method. The closest test found is a seller's blog covering one device, one 113-word passage and one speaker, with no noise levels. Kafle and Huenerfauth found in 2017 that word error rate alone has little correlation with how well deaf users understand captions, so user ratings are needed too.
hearingtracker.comhelp.xanderglasses.comdymesty.comsigaccess.org
Built with what exists
Three or more devices, such as Meta Ray-Ban Display with its captions, XanderGlasses, AirCaps and Even Realities G2; scripted speech recorded by consenting speakers and played from a loudspeaker at a set distance; recorded café and street noise played at set levels, checked with a calibrated sound level meter; reference transcripts, with OpenAI's open Whisper model and a current speech-to-text service such as ElevenLabs Scribe v3 as baselines (see the Note "Fewer word errors").
First step
Borrow or buy three devices and a sound level meter, and record 30 minutes of scripted speech from consenting volunteers.
Done looks like
A published table of word error rate and delay for at least 3 devices, 3 noise conditions and 3 speakers, with raw audio, transcripts and method released, and ratings from at least 10 deaf or hard-of-hearing testers.
Open 02 · the brief
Seeing the recording light
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A controlled test of whether people notice the recording light on camera glasses at 1, 2, 4 and 8 metres and several angles, indoors and outdoors in daylight, with the light level measured, and a published tally.
Why it is still open
Regulators have asked for field testing, and none has been published. Ireland's Data Protection Commission said in 2021 that comprehensive field testing had not been shown to it. Press reports of the Hamburg authority's 2026 findings give no count of observers. A 2026 paper for the CHI conference asked bystanders their views, and 41.3 percent rated the lights sufficient, but it measured no detection. A 2025 Virginia Tech study tested no lights.
Built with what exists
Ray-Ban Meta and Oakley Meta glasses and, once they ship, Android XR glasses; a lux meter; marked distances; a random schedule of recording and not recording. Everyone in view is a consenting participant, and footage is deleted after each trial.
First step
Write the test plan and consent form and submit them for ethics review.
Done looks like
Detection rates by distance, angle and light level from at least 30 participants, with the test plan registered in advance.
Open 03 · the brief
Describing scenes for blind users
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of AI scene description on camera glasses with blind and low-vision participants, on a fixed set of 200 real scenes such as signs, labels, door numbers and transit boards, scored by category. Money-handling tasks are left out.
Why it is still open
The one published accuracy test found, by Medina and colleagues in JAMA Ophthalmology (2026), had 6 of the study's own authors use Ray-Ban Meta glasses on objects on a tabletop. Objects were named correctly in 699 of 700 trials, standard text in 59 of 100 and coins in 3 of 150, and the authors call for studies with people who have low or no vision. Be My Eyes, which runs on the glasses, cites no evaluation of its glasses service.
Built with what exists
Ray-Ban Meta glasses with Meta AI; Meta's Wearables Device Access Toolkit, a developer preview that lets apps use the glasses' camera; a scene list and scoring rubric; a blindness organisation as partner.
First step
Draft the scene list and rubric and approach a local blindness organisation.
Done looks like
Accuracy by category for at least 10 blind or low-vision participants on 200 scenes, compared with the JAMA study's results from its own authors, with scenes and scores published.