MemetikEdition 2026-09

ResearchEdition 2026-09

OpenAI, Anthropic, Google and Perplexity agree on the leader in 24 of 33 software categories

Across ten current models, the four providers named the same most-frequent vendor in 24 categories. Where they split, it is usually one provider's one model.

Figure 01 · Edition 2026-09

AI coding agents

How often each product was named, and how often it appeared first.

NamedNamed first
  1. Cursor
    Named: 98%
    Named first: 48%
  2. Claude Code
    Named: 98%
    Named first: 38%
  3. GitHub Copilot
    Named: 98%
    Named first: 4%
  4. Codex
    Named: 78%
    Named first: 10%
  5. Cline
    Named: 60%
    Named first: 0%
  6. Aider
    Named: 52%
    Named first: 0%
  7. Windsurf
    Named: 42%
    Named first: 0%
  8. Devin
    Named: 34%
    Named first: 0%

Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09

Download source data

Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.

Key findings

  1. All four providers (three OpenAI models, three Anthropic, two Google, two Perplexity) named the same most-frequent vendor in 24 of 33 categories.

  2. All ten individual models agreed in 21 of 33. In 18 categories the leader was named in every one of the 50 answers.

  3. OpenAI was the odd provider out in 3 categories: GitHub Copilot over Cursor (coding agents), AWS RDS over Supabase (databases), Zapier over Lindy (agent builders).

  4. Within providers, one model often breaks ranks: Claude Sonnet 5 named Hostinger and AWS where the other nine named Wix and Vercel; GPT-5.6 Sol named Claude Code and Modal where Terra and Luna named Cursor and Hugging Face.

  5. The nine disagreements sit in the newest categories: agent builders, coding agents, image generation, web search APIs, browser agents, inference hosting.

Ten models from four providers agree on the most-named vendor in 24 of 33 categories. In 18 of those, the leader appeared in all 50 answers. Where the providers split, it is rarely a clean OpenAI-versus-Anthropic divide: it is usually one model within one provider breaking from the other nine.

Where do the providers disagree?

“Most named” per provider is the vendor that appeared in the largest number of that provider’s answers (15 for OpenAI and Anthropic, 10 for Google and Perplexity).

Category OpenAI Anthropic Google Perplexity
AI coding agents GitHub Copilot Cursor Cursor Cursor
Databases for startups AWS RDS Supabase Supabase Supabase
AI agent builders Zapier Lindy n8n Lindy
AI image generation Adobe Firefly Midjourney Adobe Firefly Adobe Firefly
Web search APIs for AI Exa Firecrawl Exa Exa
Browser automation Playwright Stagehand Playwright Browser Use
AI website builders Wix Wix Hostinger Wix
Creator monetization Podia Podia Kajabi Podia
AI music generation Suno Suno Suno Soundraw

Which individual models break from their provider?

Twelve categories had at least one dissenting model. The dissents cluster on a few models:

  • Claude Sonnet 5 alone named Hostinger (website builders), AWS (hosting) and GPT Image (image generation) where the other nine models named Wix, Vercel and Adobe Firefly or Midjourney.
  • GPT-5.6 Sol alone named Claude Code (coding agents), Modal (inference hosting) and n8n (agent builders).
  • Claude Fable 5 alone named Stagehand (browser agents) and Soundraw (music).
  • Sonar Pro alone named CapCut (clipping). Sonar Reasoning Pro alone named Browser Use (browser agents).
  • Gemini 3.5 Flash alone named Payhip (creator monetization) and joined Terra on Zapier (agent builders).

Where does everyone agree?

Twenty-one categories, and in eighteen of them the leader was named in every answer: HeyGen, CodeRabbit, Apollo, Veo, ElevenLabs, Kajabi, Linktree, OpenAI, LangSmith, Substack, Zapier, Circle, Amplitude, Stripe Billing, HubSpot, Gusto, Resend and Lovable. When a category has a default, ten models from four labs know it.

What the disagreements have in common

The nine provider-level splits are in the categories that did not exist or were being reset two years ago: agent builders, coding agents, browser agents, AI image generation, web search APIs for LLMs, inference hosting. Where the category is old, the models read the same web and reach the same name. Where it is new, the sources are thinner and which pages each model retrieved decides the answer.

What this means for a vendor

A vendor that leads across providers but trails in one model has a retrieval gap, not an awareness gap. The fix is specific: the pages that one model reads. A vendor that leads in one provider only has a bigger problem, and the per-provider table on its category page shows exactly where.

Limitations

“Most named” per provider rests on 10 or 15 answers; per model on 5. Ties were broken by order in the vendor list. API models with web search are close to but not identical to the consumer chat products. One edition.

Explore 1 more category chart

Figure 02 · Edition 2026-09

Databases for startups

How often each product was named, and how often it appeared first.

NamedNamed first
  1. Supabase
    Named: 98%
    Named first: 52%
  2. AWS RDS
    Named: 92%
    Named first: 24%
  3. Neon
    Named: 90%
    Named first: 22%
  4. Google Cloud SQL
    Named: 66%
    Named first: 0%
  5. PlanetScale
    Named: 64%
    Named first: 0%
  6. MongoDB
    Named: 64%
    Named first: 0%
  7. Turso
    Named: 32%
    Named first: 0%
  8. Firebase
    Named: 24%
    Named first: 0%

Top 8 by answer share · 50 answers
10 models · 5 prompts · Memetik Index, 2026-09

Download source data

Shares use all 50 answers as the denominator. Multiple products can appear in one answer. First position does not measure recommendation strength.

Behind the figures

Method and sources

Edition 2026-09: 5 fixed buyer prompts per category, sent to 10 models through their APIs. Each category contains 50 recorded answers. Answer share is the fraction of answers naming a vendor. Named first records the vendor's position before other tracked vendors, not the strength of a recommendation. This is one dated sample; small differences can reflect answer variation. API responses can differ from consumer chat products. Read the full method.

Cite this research

Memetik. OpenAI, Anthropic, Google and Perplexity agree on the leader in 24 of 33 software categories. Memetik, 2 September 2026. Data: edition 2026-09. https://www.memetik.ai/research/ai-providers-agree-on-leader-24-of-33-categories

Charts and figures: CC BY 4.0. Credit Memetik Index and link to the source.

Memetik

Memetik measures which software products AI models name when buyers ask. Its research reports connect the findings to the recorded answers.