Independent rankings of AI models, apps and tools.
Ranked by consensus from benchmarks, expert reviews and real-world testing, with the evidence behind every call.
AI models
Best AI models, October 2026
| Rank | Model | Provider | Released | Price in / out | Change |
|---|---|---|---|---|---|
| 1 | Claude Opus 5.5 Anthropic | Anthropic | 22 Sep 2026 | $4 / $20 | – |
| 2 | Claude Sonnet 5.5 Anthropic | Anthropic | 28 Sep 2026 | $2 / $10 | – |
| 3 | Muse Wonder Full Meta · New | Meta | 12 Oct 2026 | n/a | New |
| 4 | Claude Mythos 5.1Restricted access Anthropic · ▼ 1 | Anthropic | 1 Sep 2026 | n/a | ▼ 1 |
| 5 | Claude Fable 5.1 Anthropic · ▼ 1 | Anthropic | 1 Sep 2026 | $10 / $50 | ▼ 1 |
| 6 | GPT-6 Astra OpenAI · ▼ 1 | OpenAI | 3 Sep 2026 | $10 / $50 | ▼ 1 |
| 7 | Gemini 4 Argon Google · ▼ 1 | 30 Sep 2026 | $2 / $10 (to 30 Nov) | ▼ 1 | |
| 8 | GPT-6.1 Sol OpenAI · ▼ 1 | OpenAI | 29 Sep 2026 | $2 / $10 | ▼ 1 |
| 9 | Muse Wonder Meta · New | Meta | 12 Oct 2026 | $4 / $20 | New |
| 10 | Claude Mythos 5Restricted access Anthropic · ▼ 2 | Anthropic | 9 Jun 2026 | $10 / $50 | ▼ 2 |
Change since the previous ranking. Prices in US dollars per million tokens.
Newsletter
The Daily Debrief
The last 24 hours in AI.
Skimmable in 60 seconds.
OpenAI engineer arrested for bugging execs, Character.AI goes bankrupt An OpenAI engineer's arrest, Character.AI's bankruptcy, Gemini 4 Argon reaching subscribers, X's Grok and Cursor bundle, Anthropic ditching Git, and the labs' grip on outside safety testers. Read issue 44 →
- No. 43 13 Oct 2026 Meta ships its frontier model Muse Wonder, Torvalds codes with Claude
- No. 42 12 Oct 2026 Claude sent police a fake murder tip, Amazon eyes Decart
Explore
Browse the rankings
- Models Frontier, flagship and open-weight models, re-scored every month. 58 models →
- Apps The assistants people actually use, and who each one suits. 12 apps →
- Categories The best AI for each job, from coding to transcription. 106 categories →
- Comparisons Head-to-heads, with the numbers side by side. 12 comparisons →
How we rank
No position in a ranking is for sale.
Each ranking aggregates the public evidence, from vendor results and benchmarks to what users are saying. Where the sources disagree, the page says so.
Specific benchmarksWhat the market is saying
- 01 Vendor results Reported as the company’s own claim.
- 02 Published benchmarks Scores on standard public tests.
- 03 Independent evaluations Used over vendor figures where they exist.
- 04 Hands-on testing Using the product for real work.
- 05 Expert reviews Specialist reviewers and practitioners.
- 06 User ratings Aggregate ratings from the people using it.
No paid placementNo position is for sale. No entry is sponsored.
No affiliate linksWe earn nothing when you sign up for a tool.
No links for saleEvery external link is nofollow.
How we rank →