AI data
Updated · By llmwise, AI-assisted. Other companies' plans checked .
AI data studies: prices, limits, access and privacy
7 studies of AI plans and models. Four come from companies' dated pages: messages per $20, 17 price and limit changes, newest-model access and 11 apps' privacy terms. Three come from our test runs of September 27, 28, 29 and October 2, 7, 8, 9, 15, 2026: 20 models on 50 prompts, ranked by passes and cost per finished task.
What did each study find?
Each study opens on its finding. Here they are together, with a link to the numbers behind each.
Messages per $20: every AI plan that publishes a count, and llmwise's: On llmwise Pro ($20), $20 buys up to 31 GPT-6 Astra messages a month, up to 125 on Claude Sonnet 5, and 60 everyday messages a day. Of 13 AI subscriptions we track, none publishes a chat count, so the table sets llmwise beside the chat apps that publish one.
AI price and limit changelog: every dated change, sourced: 17 dated changes to AI plans, offers and API prices happened between January 16, 2026 and October 13, 2026, each from the company's own page. 7 changed plans' prices, limits or models, 2 offers ended, and the rest were launches and API prices. The RSS feed carries new ones.
Which AI plan gets the newest models? Plan by plan: GPT-6 Astra is in the regular chat of 2 of the 5 plans whose pages name it; GPT-6 Sol is in the regular chat of 1 of the 5 plans whose pages name it. On llmwise Pro ($20), all 12 of these newest models that llmwise carries are in the chat, each with its count shown before you send.
AI chat privacy scorecard: who trains on your chats, and for how long they keep them: Of 11 AI chat apps, 7 train on your chats unless you opt out; Claude only if you allow it; Meta AI names no opt-out on the page we read; llmwise and Poe don't. Where stated, chats are kept from 30 days to 18 months, longer for some.
Price per task: what finishing an AI task costs, model by model: Finishing one task cost from $0.00013 on GPT-6 Luna to $0.0223 on Claude Fable 5.1, 177 times as much, in our test runs of September 27, 28, 29 and October 2, 7, 8, 9, 15, 2026. A finished task is a reply that passed its check. Each figure is OpenRouter's charge for all 50 prompts, divided by how many passed.
AI model quality tracker: the same prompts, every model, dated: A dated snapshot, not a trend yet: all 20 models ran the same 50 prompts on September 27, 28, 29 and October 2, 7, 8, 9, 15, 2026. They passed from 40 (GLM 5.3 Flash) to 49 (DeepSeek V4 Pro); all passed the same 33. We rerun within 45 days, or when a price or version changes, keeping each run.
The AI model leaderboard, job by job: Each job's leaderboard from our test runs of September 27, 28, 29 and October 2, 7, 8, 9, 15, 2026: all 20 models ranked on its five prompts by passes, hard passes, then message size and cost. The top of the table runs from 3 models sharing first on writing to 20 sharing it on RAG and answering from documents.
The studies
- Messages per $20: every AI plan that publishes a count, and llmwise's
Messages per $20 on every llmwise plan and model, beside every AI plan that publishes a count (the most: ChatHub Pro (yearly)), and which don't.
- AI price and limit changelog: every dated change, sourced
17 dated changes to AI subscriptions' prices, limits and models, offers that ended and API price moves, each sourced. With an RSS feed.
- Which AI plan gets the newest models? Plan by plan
Which AI plan has GPT-6 Astra and the other newest models in its regular chat, plan by plan in the companies' words, beside llmwise.
- AI chat privacy scorecard: who trains on your chats, and for how long they keep them
Does ChatGPT, Claude, Gemini, Copilot or Grok train on your chats? 11 apps' training, opt-out and retention, in their own words.
- Price per task: what finishing an AI task costs, model by model
What one AI task costs, not per token: 20 models' price per task from our test runs, $0.00013 to $0.0223, job by job.
- AI model quality tracker: the same prompts, every model, dated
Did they nerf it? 20 AI models on the same 50 prompts, dated, with the version that answered: a snapshot we run again every 45 days.
- The AI model leaderboard, job by job
The AI model leaderboard, job by job: all 20 models ranked on 10 jobs by our own scored tests, with what each reply cost and how long it took.
- AI usage limits: every $20 plan's limits, explained
Each $20 AI plan's limits in its company's words, and llmwise's counts.
- Subscription stack calculator
Tick the AI subscriptions you pay for: the total, and what one llmwise plan gives.
Also from primary sources: OpenRouter's usage accounting docs (the cost OpenRouter reports for every request, which these figures add up). Anthropic's Claude Opus 5.5 page (the model that grades the rubric prompts). OpenAI's GPT-6 Astra docs (the model that grades Claude Opus 5.5's own replies). Read .
Our data: questions
Where does the data come from?
From each company's own pages, quoted and dated, and from llmwise's plans for llmwise's numbers. Nothing is estimated: where a company publishes no figure, the study says so.
Can I cite these?
Yes, with a link. The changelog also has an RSS feed at /data/changelog.xml.
The numbers here, in the app
The counts and prices behind these studies are the app's own: 20 models, each priced per message before you send.