Which model you actually get: a sourced map of AI assistant lineups (October 2026)
By Bot Tutorials · Published Oct 29, 2026 · Updated Oct 29, 2026 · Facts last verified Oct 29, 2026
A cited comparison of the models behind ChatGPT, Claude, Gemini, Perplexity and GitHub Copilot as of Oct 29, 2026. It covers which model runs where, how vendors present their claims, and what we couldn't verify.
1. Question and method
We wanted to know which model a user actually gets in each product, and through which surface. We read only official announcements, release notes and changelogs on Oct 29, 2026. Each row below gives its source. Vendor benchmark figures are reported as claims and never compared across vendors, because each vendor uses different tests and settings.
2. Findings
| Product / surface | Model(s) per official source | Source |
|---|---|---|
| ChatGPT Chat tab, paid | GPT-6 Sol (Intelligent UI), from Oct 7, 2026 | OpenAI |
| ChatGPT Chat tab, Free and Go | GPT-6 Luna | OpenAI |
| ChatGPT Work and Codex | GPT-6.1 Terra is selectable from Oct 26, 2026. GPT-6.1 Sol and GPT-6 Astra are also named | OpenAI |
| Claude API, small model | claude-haiku-5-5 (Oct 7, 2026) | Anthropic |
| Claude, frontier | Opus 5.5 (Sep 22, 2026) and Sonnet 5.5 listed in the newsroom | Anthropic |
| Gemini API Flash | gemini-3.6-flash and gemini-3.8-flash. Old 3.5 and 3.7 strings are rerouted | |
| Perplexity Computer | GPT-6 Astra, GPT-6.1 Sol (Ultrafast for Max), Claude Opus 5.5, and local Qwen 3.6 or PPLX Pro on Mac | Perplexity |
| GitHub Copilot | Copilot Fusion routing, plus Claude Haiku 5.5, GPT-6.1 Terra and Gemini 4 Argon added in October | GitHub changelog |
3. Analysis
- The product name doesn't tell you the model. In ChatGPT the model depends on both the plan and the tab (Chat vs Work or Codex).
- Routing is becoming the default. Copilot Fusion and Gemini's alias rerouting both mean the model behind a request can change without you changing anything. For anything repeatable, pin a model string where you can and rerun your test set when releases come out.
- Cost claims are relative. “50% lower than GPT-5.6 Terra” or “75% less than Haiku 4.5” compare against an older model, not against competitors.
4. Limitations
Consumer price tables for ChatGPT, Gemini and Perplexity weren't readable as text, and several OpenAI help pages blocked our checker, so this paper leaves prices out. Gemini app (consumer) model assignments aren't covered because we didn't find an October 2026 official source.