Book a demo

AI model leaderboard

See which AI models brands use for customer support on Richpanel. Compare usage and adoption.

Book a demo Download JSONAbout this data
674Btokens, last 3 months
11.9MAI conversations, last 3 months

Ranked by usage. These charts do not measure answer quality. Anonymized, through Sep 29, 2026.

Token volume

Share of Richpanel token volume across the top models.

  1. 1Gemini 3 FlashGoogle48.4%
  2. 2GPT-5.6 LunaOpenAI28.4%
  3. 3Gemini 2.5 FlashGoogle11.4%
  4. 4Sonnet 5Anthropic3.6%
  5. 5GPT-6 LunaOpenAI3.4%
  6. 6Sonnet 4.6Anthropic2.6%
  7. 7GPT-5.4OpenAI0.7%
  8. 8Opus 5Anthropic0.4%
  9. 9Other1.2%

Conversations

Share of AI conversations across the top models, all work types.

  1. 1GPT-5.6 LunaOpenAI48.5%
  2. 2Gemini 2.5 FlashGoogle24.2%
  3. 3Gemini 3 FlashGoogle17.8%
  4. 4GPT-6 LunaOpenAI6.9%
  5. 5Sonnet 4.6Anthropic1.0%
  6. 6Sonnet 5Anthropic0.9%
  7. 7GPT-5.4OpenAI0.2%
  8. 8grok-4.3xAI0.2%
  9. 9Other0.3%

Cost per conversation

What brands pay per conversation the AI resolves with no human reply, by model. Credits at $0.50, the annual rate.

  1. 1GPT-5.6 LunaOpenAI$0.140.28 credits
  2. 2GPT-6 LunaOpenAI$0.250.50 credits
  3. 3Gemini 3 FlashGoogle$0.270.54 credits
  4. 4Kimi K2.5Moonshot$0.350.69 credits
  5. 5Gemini 2.5 FlashGoogle$0.350.70 credits
  6. 6Sonnet 4.6Anthropic$0.480.95 credits
  7. 7GPT-5.4OpenAI$0.561.11 credits
  8. 8Sonnet 5Anthropic$0.811.63 credits

Reach

Share of brands using each model, week by week. Brands use several models, so these shares do not sum to 100%.

  1. 1GPT-5.6 LunaOpenAI92.6%
  2. 2Gemini 3 FlashGoogle91.3%
  3. 3GPT-6 LunaOpenAI88.8%
  4. 4Gemini 2.5 FlashGoogle69.4%
  5. 5GPT-5.6 SolOpenAI45.9%
  6. 6Sonnet 5Anthropic24.8%
  7. 7GPT-6 SolOpenAI24.0%
  8. 8Opus 5Anthropic12.8%

Preference

Share of brands with each model as their primary model, by token volume.

  1. 1Gemini 3 FlashGoogle34.4%
  2. 2GPT-5.6 SolOpenAI24.0%
  3. 3GPT-5.6 LunaOpenAI15.0%
  4. 4Sonnet 5Anthropic10.2%
  5. 5Gemini 2.5 FlashGoogle5.1%
  6. 6GPT-6 SolOpenAI3.6%
  7. 7Opus 5Anthropic3.3%
  8. 8Sonnet 4.6Anthropic2.3%

Questions about usage and cost.

Gemini 3 Flash carries 48% of token volume across brands on Richpanel in the last 3 months, then GPT-5.6 Luna at 28%. This ranks usage, not answer quality.

No. Usage says which model brands chose, not how well it answers. Every model on this page passed Richpanel's 10,000+ pre-launch tests before it was offered. Which one a brand picks comes down to its volume, its policies and its budget.

By token volume across brands on Richpanel: Google 60%, OpenAI 33%, Anthropic 7%. Switch the leaderboard to Top labs for the full split, and to AI Manager to see which models run the strategic work.

It depends on the model. GPT-5.6 Luna resolves a conversation for about $0.14; Sonnet 5 for about $0.81. That is the credits the conversation used at $0.50 a credit, the annual rate, for conversations the AI resolved with no human reply.

The pricing page shows an approximate cost per conversation with a buffer built in. This page shows what a brand is billed for conversations the AI resolved on its own. Your bill also covers AI Manager and background work such as summaries, translation and quality checks. The pricing-page buffer accounts for those.

The unit Richpanel bills AI work in. A credit costs $0.50 on an annual plan, $0.60 billed monthly, $0.65 on demand. Each model uses a different number of credits per conversation, which is what the cost chart shows.

Yes. You pick the model and the credit budget for your workspace and can change either at any time. The models on this page are the ones brands have picked.

Those count every kind of AI traffic through a gateway. This counts customer support work only: conversations with customers, drafts for human agents, AI Manager and background tasks, across brands on Richpanel, from production usage.

Brands run more than one model. A brand on two models counts for both.

It prices only conversations the AI resolved with no human reply, on one model. Conversations where a human replied, or that used two models, are out. Background work on a conversation is not counted.

Every night, through the previous full day. Download the whole export as JSON or any chart as CSV, and link to this page when you cite it. No brand is named and no per-brand figure is published.

About this data

Data source
Actual AI use across brands on Richpanel. Internal and test workspaces are excluded.
Usage
Tokens include input, output and cached tokens. Conversations count every time an AI model works on one: a customer reply, a draft for an agent, a summary or a review each count.
Cost per conversation
What brands pay Richpanel for conversations the AI resolves with no human reply. A human approving an AI action still counts. Support AI and Agent Copilot credits at $0.50 per credit, the annual rate. Excludes background work and conversations that used more than one model. Counted on the day they are resolved.
Work included
Support AI answering customers, Agent Copilot working beside your team, AI Manager and background tasks. Background tasks include summaries, translation, quality checks and reports. The work-type filter applies to token volume, reach and preference.
How brands are counted
Each brand counts once per model each week. Its most-used model handles the largest share of its tokens during the selected period.

Brand names and individual usage stay private. Download all data as JSON, or each chart as CSV.