The rankings
Which AI is best at what
Every list below is calculated from independent benchmarks, not from opinions or advertising. A subscription gets the score of the best AI model it includes. Last updated 26 Oct 2026.
Best AI overall
The best all-rounder for everyday questions, writing and thinking. Read the verdict →
- 98based on 4 of 5 tests
- 97based on 4 of 5 tests
- 92based on 4 of 5 tests
- 83based on 3 of 5 tests
- 82based on 5 of 5 tests
- 72based on 3 of 5 tests
- 71based on 5 of 5 tests
- 55based on 2 of 5 tests
Best free: Claude Free · Best value (paid): Claude Pro
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: overall chat (40%): Millions of people compare two anonymous AI answers and pick the better one. Data checked 26 Oct 2026.
- Artificial Analysis Intelligence Index (27%): A combined score over many independent tests of knowledge and reasoning. Data checked 19 Oct 2026.
- GPQA Diamond (13%): PhD-level science questions that can't simply be googled. Data checked 26 Oct 2026.
- SimpleQA Verified (13%): Short factual questions: does the AI know the answer or make something up? Data checked 19 Oct 2026.
- LiveBench: reasoning (7%): Logic puzzles and spatial reasoning. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Benchmark data: Artificial Analysis (artificialanalysis.ai) · Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0 · Benchmark data: LiveBench (livebench.ai)
Best AI for coding
Writing, fixing and explaining code, and building websites and apps. Read the verdict →
- 99based on 2 of 6 tests
- 99based on 2 of 6 tests
- 99based on 2 of 6 tests
- 94based on 3 of 6 tests
- 82based on 2 of 6 tests
- 72based on 2 of 6 tests
- 71based on 6 of 6 tests
- 71based on 2 of 6 tests
- 50based on 3 of 6 tests
Best free: GitHub Copilot Free · Best value (paid): GitHub Copilot Pro
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: coding questions (14%): Blind votes on answers to programming questions. Data checked 26 Oct 2026.
- Arena: building websites (14%): Blind votes on websites and apps the AI builds from a prompt. Data checked 26 Oct 2026.
- SWE-bench Verified (21%): Can the AI fix real bugs in real open-source software projects? Data checked 26 Oct 2026.
- Artificial Analysis Coding Index (21%): A combined score over independent coding tests. Data checked 26 Oct 2026.
- LiveBench: coding (14%): Fresh programming problems that models cannot have memorised. Data checked 26 Oct 2026.
- LiveBench: agentic coding (14%): Multi-step coding tasks in Python, JavaScript and TypeScript projects. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0 · Benchmark data: Artificial Analysis (artificialanalysis.ai) · Benchmark data: LiveBench (livebench.ai)
Best AI for writing
Emails, essays, stories, editing and rewriting. Read the verdict →
- 100based on 2 of 3 tests
- 98based on 2 of 3 tests
- 87based on 2 of 3 tests
- 84based on 2 of 3 tests
- 75based on 2 of 3 tests
- 72based on 3 of 3 tests
- 67based on 2 of 3 tests
Best free: Claude Free · Best value (paid): Claude Pro
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: creative writing (38%): Blind votes on stories, poems and other creative texts. Data checked 26 Oct 2026.
- Arena: writing & language (38%): Blind votes on everyday writing, editing and language tasks. Data checked 26 Oct 2026.
- LiveBench: following instructions (25%): Summarising, simplifying and rewriting text exactly as asked. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Benchmark data: LiveBench (livebench.ai)
Best AI for research
Finding facts with sources, searching the web and writing research reports.
- 98based on 2 of 3 tests
- 90based on 2 of 3 tests
- 89based on 3 of 3 tests
- 89based on 3 of 3 tests
Best free: ChatGPT Free · Best value (paid): ChatGPT Plus
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: search & research (43%): Blind votes on answers that use live web search. Data checked 15 Oct 2026.
- SimpleQA Verified (29%): Short factual questions: does the AI know the answer or make something up? Data checked 19 Oct 2026.
- DeepResearch Bench (29%): Writing well-sourced research reports on complex questions. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0
Best AI for maths and hard problems
Maths, science and problems that need careful step-by-step thinking.
- 98based on 3 of 5 tests
- 97based on 3 of 5 tests
- 95based on 3 of 5 tests
- 89based on 4 of 5 tests
- 86based on 3 of 5 tests
- 79based on 2 of 5 tests
- 78based on 5 of 5 tests
Best free: Claude Free · Best value (paid): ChatGPT Plus
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: maths (20%): Blind votes on answers to maths questions. Data checked 26 Oct 2026.
- GPQA Diamond (20%): PhD-level science questions that can't simply be googled. Data checked 26 Oct 2026.
- FrontierMath (tiers 1-3) (20%): Very hard, unpublished maths problems written by mathematicians. Data checked 26 Oct 2026.
- LiveBench: maths (20%): Recent competition maths and hard calculation problems. Data checked 26 Oct 2026.
- LiveBench: reasoning (20%): Logic puzzles and spatial reasoning. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks, CC BY 4.0 · Benchmark data: LiveBench (livebench.ai)
Best AI for students
Explaining material, homework help, maths and writing feedback. Read the verdict →
- 100based on 3 of 5 tests
- 98based on 4 of 5 tests
- 94based on 4 of 5 tests
- 89based on 5 of 5 tests
- 89based on 3 of 5 tests
- 77based on 4 of 5 tests
- 77based on 4 of 5 tests
- 67based on 4 of 5 tests
Best free: Claude Free · Best value (paid): Gemini Google AI Pro
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: homework help (38%): Blind votes on help with homework from photos and screenshots. Data checked 26 Oct 2026.
- Arena: overall chat (25%): Millions of people compare two anonymous AI answers and pick the better one. Data checked 26 Oct 2026.
- Arena: maths (13%): Blind votes on answers to maths questions. Data checked 26 Oct 2026.
- Arena: writing & language (13%): Blind votes on everyday writing, editing and language tasks. Data checked 26 Oct 2026.
- LiveBench: maths (13%): Recent competition maths and hard calculation problems. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Benchmark data: LiveBench (livebench.ai)
Best AI for business
Business questions, documents, spreadsheets and following instructions precisely. Read the verdict →
- 100based on 2 of 5 tests
- 98based on 2 of 5 tests
- 94based on 3 of 5 tests
- 89based on 2 of 5 tests
- 81based on 3 of 5 tests
- 78based on 4 of 5 tests
- 75based on 2 of 5 tests
Best free: Claude Free · Best value (paid): Gemini Google AI Pro
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: business & finance (22%): Blind votes on business, management and finance questions. Data checked 26 Oct 2026.
- Arena: working with documents (22%): Blind votes on answers about uploaded documents. Data checked 26 Oct 2026.
- Arena: following instructions (22%): Blind votes on how well the AI does exactly what it was asked. Data checked 26 Oct 2026.
- LiveBench: data analysis (22%): Working with tables: joining, reformatting and finding patterns. Data checked 26 Oct 2026.
- LiveBench: following instructions (11%): Summarising, simplifying and rewriting text exactly as asked. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Benchmark data: LiveBench (livebench.ai)
Best AI for images
Creating images from a description and editing existing images. Read the verdict →
Not enough verified data to rank subscriptions for this topic yet.
Best free: not enough data · Best value (paid): not enough data
How this list is made
Scores combine these independent tests (weights in brackets). 100 = the best model from the companies we cover. A subscription gets the score of the best model it includes. Full method.
- Arena: image generation (40%): Blind votes on images made from a text prompt. Data checked 26 Oct 2026.
- Arena: image editing (20%): Blind votes on edits the AI makes to an existing image. Data checked 26 Oct 2026.
- Artificial Analysis: image arena (40%): Blind votes on images made from a text prompt. Data checked 26 Oct 2026.
Leaderboard data: Arena (arena.ai), CC BY 4.0 · Benchmark data: Artificial Analysis (artificialanalysis.ai)