Frontier models

Models

Who the AI conversation is about, lab by lab. Each lab has its own reading every week, which counts every post about the lab, its model families and its flagship products once. Below it: each lab’s families and products on the platforms’ summary tables, what moved week to week, and the full lineage map.

12labs
37model families
48flagship products
3,858readings, all in development
40weeks read: W01–W40
Not yetversions: not in the registry yet

Counts from the stored readings and the stored releases 2026-W01–W40, as of 8 Oct 2026 12:00Z. In development

Labs this week

Each lab’s own reading for 28 Sep – 4 Oct: how many distinct people posted about the lab, its model families or its flagship products, counted once per post. Labs are compared here only with labs.

A lab’s figure is not the sum of its subjects: a post that names two of its subjects counts once for the lab. That is why a lab’s order here can differ from its families’ order on the summary tables.

Platform

In development

LabDistinct people on Redditscaled to the top lab here, Anthropic at 12,518Peoplen · coverage · the reading

Anthropic12,518n 19,858 posts · author coverage 99.9%r_0af67f21aa57

OpenAI10,505n 15,494 posts · author coverage 99.9%r_18055a6a2ae4

Google DeepMind3,074n 4,081 posts · author coverage 100.0%r_ffec06a32699

Alibaba2,258n 3,501 posts · author coverage 100.0%r_cbd17342ec31

DeepSeek1,136n 1,503 posts · author coverage 99.9%r_a3976cc40196

Meta1,007n 1,238 posts · author coverage 100.0%r_9a8417d6a83a

xAI1,001n 1,382 posts · author coverage 100.0%r_dd608a78b33b

MiniMax814n 1,328 posts · author coverage 100.0%r_447c648bc397

Zhipu AI678n 857 posts · author coverage 100.0%r_14c234ce09be

Microsoft AI618n 795 posts · author coverage 99.9%r_97020043e9e5

Moonshot AI232n 288 posts · author coverage 100.0%r_a85acbb1fe24

Mistral AI169n 200 posts · author coverage 100.0%r_1578bda1dade

Distinct people who posted about each lab, its families or its products, on Reddit, 28 Sep – 4 Oct 2026 (2026-W40), as of 8 Oct 2026 12:00Z; in development. Each lab’s figure is from its own W40 reading and counts each post once across its families and products, so it is never added to, or drawn beside, its own subjects’ figures. Each bar starts at zero on this chart’s own scale, where the top lab is the full bar; n is posts (n 200 to 19,858 per lab). Reddit 7 days: 7 active; author coverage 99.9%–100.0%. All 12 labs have a W40 reading. Labs are sorted by the figure shown: no stored object ranks labs.

The same labs on three more platforms, each on its own scale

Bluesky

12 labs, 1 hidden · full bar OpenAI, 3,444

7 days: 7 active · author coverage 100.0%

OpenAI3,444 · n 6,511
Anthropic2,760 · n 4,543
Google DeepMind1,182 · n 1,800
xAI621 · n 764
Meta498 · n 712
DeepSeek351 · n 486
Microsoft AI342 · n 509
Alibaba332 · n 522
Moonshot AI140 · n 170
Zhipu AI131 · n 177
Mistral AI126 · n 144
MiniMax
Hidden. n 18 people, below 20

YouTube

12 labs, 6 hidden · full bar OpenAI, 610

7 days: 7 active · author coverage 100.0%

OpenAI610 · n 680
Anthropic305 · n 331
Google DeepMind56 · n 63
Meta52 · n 52
xAI46 · n 46
Microsoft AI20 · n 21
Alibaba
Hidden. n 10 people, below 20
DeepSeek
Hidden. n 9 people, below 20
Moonshot AI
Hidden. n 6 people, below 20
Mistral AI
Hidden. n 3 people, below 20
MiniMax
Hidden. n 1 people, below 20
Zhipu AI
Hidden. n 1 people, below 20

Hacker News

12 labs, 10 hidden · full bar OpenAI, 41

7 days: 7 active · author coverage 100.0%

OpenAI41 · n 42
Anthropic25 · n 25
Google DeepMind
Hidden. n 7 people, below 20
Meta
Hidden. n 6 people, below 20
Alibaba
Hidden. n 2 people, below 20
DeepSeek
Hidden. n 2 people, below 20
xAI
Hidden. n 2 people, below 20
Zhipu AI
Hidden. n 2 people, below 20
Microsoft AI
Hidden. n 1 people, below 20
Mistral AI
Hidden. n 1 people, below 20
MiniMax
Hidden. n 0 people, below 20
Moonshot AI
Hidden. n 0 people, below 20

Hidden: 1 of the 12 labs on Bluesky, 6 on YouTube and 10 on Hacker News, none on Reddit. A figure under 20 people is hidden, never shown as a number, and each hatched box carries the stored reason, shortened on a narrow row (“n 7 people, below 20”). Some hidden rows are n 0: the lab reading exists, with no linked posts on that platform. Nothing is added across platforms, because one person can post on several. arXiv and Exa carry articles, which have no authors, so they have no people figure.

Release 2026-W40 How a post is linked to a lab

The twelve labs

One card per lab: its readings, its three biggest families or products on Reddit this week, and where one of them tops a platform’s table.

a reading that week, linked
no reading
in order · W33 · W34 · W35 · W36

In development

Earlier weeks

W34–W36 of 40 stored weeks

Later weeks

Earlier weeks

W33–W36 of 40 stored weeks

Later weeks

Alibaba
3 families · 4 products · 317 readings
its own readings
Top on Reddit · W40
- #5Qwen family2,170 · n 3,328
- #20Wan family122 · n 188
- –Qwen Studio product
  Hidden. n 4 people, below 20
4 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Anthropic
1 family · 4 products · 240 readings
its own readings
Top on Reddit · W40
- #1Claude family10,876 · n 16,896
- #7Claude Code product1,686 · n 2,198
- #16Claude Cowork product225 · n 282
2 more of its subjects on Reddit’s table
Leads in W40: Claude is #1 on Reddit (of 85, 44 hidden) and Bluesky (of 85, 60 hidden).
In December: Claude #1 on Hacker News (W01, of 85, 84 hidden).
In January: Claude #1 on Hacker News (W02, of 85, 83 hidden; W03, of 85, 83 hidden; W04, of 85, 83 hidden; W05, of 85, 84 hidden).
In February: Claude #1 on Reddit (W06, of 85; W07, of 85; W08, of 85; W09, of 85) and Hacker News (W06, of 85, 83 hidden; W07, of 85, 83 hidden; W08, of 85, 83 hidden; W09, of 85, 83 hidden).
In March: Claude #1 on Reddit (W10, of 85; W11, of 85; W12, of 85; W13, of 85; W14, of 85) and Hacker News (W10, of 85, 83 hidden; W11, of 85, 83 hidden; W12, of 85, 83 hidden; W13, of 85, 83 hidden; W14, of 85, 83 hidden).
In April: Claude #1 on Reddit (W15, of 85; W16, of 85; W17, of 85; W18, of 85) and Hacker News (W15, of 85, 83 hidden; W16, of 85, 83 hidden; W17, of 85, 83 hidden; W18, of 85, 83 hidden).
In May: Claude #1 on Reddit (W19, of 85, 45 hidden; W20, of 85, 46 hidden; W21, of 85; W22, of 85, 43 hidden) and Hacker News (W19, of 85, 84 hidden; W20, of 85, 84 hidden; W21, of 85, 84 hidden; W22, of 85, 84 hidden).
In June: Claude #1 on Reddit (W23, of 85; W24, of 85, 43 hidden; W25, of 85, 44 hidden; W26, of 85, 45 hidden; W27, of 85) and Hacker News (W24, of 85, 84 hidden; W25, of 85, 84 hidden; W26, of 85, 84 hidden; W27, of 85, 84 hidden).
In July: Claude #1 on Reddit (W28, of 85, 43 hidden; W29, of 85; W30, of 85, 44 hidden; W31, of 85, 43 hidden) and Hacker News (W28, of 85, 84 hidden; W29, of 85, 84 hidden; W30, of 85, 83 hidden; W31, of 85, 84 hidden).
In August: Claude #1 on Reddit (W32, of 85; W33, of 85; W34, of 85, 43 hidden; W35, of 85, 43 hidden; W36, of 85, 43 hidden) and Hacker News (W36, of 85, 84 hidden).
In September: Claude #1 on Reddit (W37, of 85, 43 hidden; W38, of 85, 43 hidden; W39, of 85, 45 hidden), Bluesky (W38, of 85, 64 hidden; W39, of 85, 59 hidden), YouTube (W38, of 85, 83 hidden; W39, of 85, 80 hidden) and Hacker News (W37, of 85, 84 hidden; W38, of 85, 84 hidden; W39, of 85, 84 hidden).
Lab page
DeepSeek
1 family · 3 products · 200 readings
its own readings
Top on Reddit · W40
- #8DeepSeek family1,136 · n 1,503
- #24DeepSeek Harness product77 · n 84
- #30DeepSeek API product47 · n 49
1 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Google DeepMind
3 families · 10 products · 560 readings
its own readings
Top on Reddit · W40
- #3Gemini family2,458 · n 3,209
- #14Gemma family368 · n 501
- #21Google Antigravity product119 · n 128
10 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Meta
2 families · 3 products · 238 readings
its own readings
Top on Reddit · W40
- #10Muse family739 · n 890
- #15Llama family243 · n 303
- #29Muse (agent app) product52 · n 57
2 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Microsoft AI
2 families · 3 products · 240 readings
its own readings
Top on Reddit · W40
- #13Microsoft Copilot product480 · n 629
- #22GitHub Copilot product118 · n 128
- –MAI family
  Hidden. n 10 people, below 20
2 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
MiniMax
4 families · 5 products · 398 readings
its own readings
Top on Reddit · W40
- #12MiniMax H-series family598 · n 934
- #41MiniMax M-series family22 · n 24
- –Hailuo family
  Hidden. n 11 people, below 20
6 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Mistral AI
9 families · 3 products · 509 readings
its own readings
Top on Reddit · W40
- #33Mistral family40 · n 44
- –Ministral family
  Hidden. n 9 people, below 20
- –Leanstral family
  Hidden. n 4 people, below 20
9 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Moonshot AI
1 family · 3 products · 199 readings
its own readings
Top on Reddit · W40
- #17Kimi family224 · n 278
- –Kimi Code product
  Hidden. n 7 people, below 20
- –Kimi app product
  Hidden. n 1 people, below 20
1 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
OpenAI
8 families · 4 products · 520 readings
its own readings
Top on Reddit · W40
- #2ChatGPT product4,786 · n 6,215
- #4GPT family2,237 · n 2,986
- #6Codex product1,947 · n 2,521
9 more of its subjects on Reddit’s table
Leads in W40: ChatGPT is #1 on YouTube (of 85, 77 hidden).
In December: ChatGPT #1 on Reddit (W01, of 85, 46 hidden).
In January: ChatGPT #1 on Reddit (W02, of 85, 44 hidden; W03, of 85, 43 hidden; W04, of 85, 43 hidden; W05, of 85).
In September: ChatGPT #1 on YouTube (W38, of 85, 83 hidden).
Lab page
xAI
1 family · 3 products · 200 readings
its own readings
Top on Reddit · W40
- #9Grok family931 · n 1,276
- #36Grok app product36 · n 40
- –Grok Build product
  Hidden. n 8 people, below 20
1 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page
Zhipu AI
2 families · 3 products · 237 readings
its own readings
Top on Reddit · W40
- #11GLM family655 · n 828
- –ZCode product
  Hidden. n 2 people, below 20
- –ChatGLM family
  Hidden. n 1 people, below 20
2 more of its subjects on Reddit’s table
No #1 on any platform’s table, in W40 or in December, January, February, March, April, May, June, July, August and September.
Lab page

Bars: distinct people on Reddit, 28 Sep – 4 Oct 2026 (2026-W40), as of 8 Oct 2026 12:00Z; in development. Each bar is on Reddit’s summary-table scale, where the table’s #1, Claude at 10,876 (n 16,896 posts), is the full bar: a different scale from the lab chart above. Rank is out of the 85 subjects on the table, hidden ones included; n is posts. A product’s posts also count in its family, so rows are never added. “Leads” names a stated #1 from the stored tables, with the table’s size; earlier weeks are listed by month. Dots link to the lab’s own reading each week.

Week to week (W39 → W40)

Families and products with at least 100 Reddit posts in both weeks, grouped by lab: people in W39, then in the week that follows, W40.

W39, 21–27 Sep 2026
W40, 28 Sep – 4 Oct 2026
one shared scale, 0 to 11,000 people

In development

SubjectW39W40

Alibaba

Qwen family2,566W39 · n 4,353 · #32,170W40 · n 3,328 · #5

Wan family95W39 · n 138 · #23122W40 · n 188 · #20

Anthropic

Claude family10,887W39 · n 16,714 · #110,876W40 · n 16,896 · #1

Claude Code product1,697W39 · n 2,089 · #71,686W40 · n 2,198 · #7

Claude app product204W39 · n 236 · #18166W40 · n 197 · #18

Claude Cowork product167W39 · n 210 · #19225W40 · n 282 · #16

DeepSeek

DeepSeek family1,297W39 · n 1,855 · #81,136W40 · n 1,503 · #8

Google DeepMind

Gemini family2,062W39 · n 2,607 · #52,458W40 · n 3,209 · #3

Gemma family446W39 · n 599 · #14368W40 · n 501 · #14

Google Antigravity product96W39 · n 108 · #22119W40 · n 128 · #21

The lineage map

Every lab with its model families and flagship products, and a dot for each week that has a reading. The map is also the index of readings.

a reading that week, linked
no reading
in order · W33 · W34 · W35 · W36

Earlier weeks

W34–W36 of 40 stored weeks

Later weeks

Earlier weeks

W33–W36 of 40 stored weeks

Later weeks

Bold: a model familyRegular: a flagship product, indented under its family

Lab › family › product, as each stored subject’s lineage names it: 37 families and 48 products under 12 labs; a product with no family hangs off its lab. Dots link to each week’s reading: W01 is 29 Dec 2025 – 4 Jan, W02 is 5–11 Jan, W03 is 12–18 Jan, W04 is 19–25 Jan, W05 is 26 Jan – 1 Feb, W06 is 2–8 Feb, W07 is 9–15 Feb, W08 is 16–22 Feb, W09 is 23 Feb – 1 Mar, W10 is 2–8 Mar, W11 is 9–15 Mar, W12 is 16–22 Mar, W13 is 23–29 Mar, W14 is 30 Mar – 5 Apr, W15 is 6–12 Apr, W16 is 13–19 Apr, W17 is 20–26 Apr, W18 is 27 Apr – 3 May, W19 is 4–10 May, W20 is 11–17 May, W21 is 18–24 May, W22 is 25–31 May, W23 is 1–7 Jun, W24 is 8–14 Jun, W25 is 15–21 Jun, W26 is 22–28 Jun, W27 is 29 Jun – 5 Jul, W28 is 6–12 Jul, W29 is 13–19 Jul, W30 is 20–26 Jul, W31 is 27 Jul – 2 Aug, W32 is 3–9 Aug, W33 is 10–16 Aug, W34 is 17–23 Aug, W35 is 24–30 Aug, W36 is 31 Aug – 6 Sep, W37 is 7–13 Sep, W38 is 14–20 Sep, W39 is 21–27 Sep and W40 is 28 Sep – 4 Oct. 3,858 readings in all, every one in development. A dotted mark between dots is where the way weeks were captured changes; the weeks run on across it.

Versions, and where next

Versions: not in the registry yet.

The map stops at families and products. A version is added with its release date, taken from two sources that don’t cite each other; until it has one, it’s held. Once versions are in, each family opens to them here.