AI Updates

LLM leaderboard and AI model rankings, with the evidence attached

Models ranked on published benchmark results, every figure verified, and every release and price change logged as it lands. The tracker shows what a ranking rests on, because a score without its coverage is a number you cannot check.

LLM leaderboard

Ranked on published benchmark results and nothing else. A model with no published score is not last, it is untested, so it does not appear here.

Full leaderboard →Every tracked model →
LLM leaderboard · 35 of 418 models have published scores
#ModelScore
1MiniMax M2.575.8
2Gemini 3 Flash Preview75.8
3Claude Opus 4.675.6
4Claude Opus 4.574.4
5GLM 572.8
6GPT-5.2-Codex72.8
7GPT-5.272.8
8GPT-571.8
9Kimi K2.570.8
10Claude Sonnet 470.8
11Claude Sonnet 4.570.6
12DeepSeek V3.270.0
13Claude Haiku 4.566.6
14GPT-5.1-Codex66.0
15GPT-5.166.0
16o164.6
17Kimi K2 Thinking63.4
18MiniMax M261.0
19Qwen3 Coder 30B A3B Instruct60.4
20GPT-5 Mini59.8
21o358.4
22GLM 4.655.4
23GLM 4.554.2
24Gemini 2.5 Pro53.6
25o4 Mini45.0

The update feed

1–20 of 534

November 202620 updates

Browse by brand

Every provider WriteWorks tracks. Each brand page carries its updates and its models, and every model has its own page underneath.

LLM leaderboards, benchmarks and rankings, explained

What the ranking above does and does not tell you.

What is an LLM leaderboard?
An LLM leaderboard ranks large language models against a fixed set of benchmarks so their results can be read side by side. The ranking is only as good as the evidence under it: a score with no published source, or a model tested on three benchmarks sitting above one tested on thirty, tells you less than it appears to.
How are these AI rankings calculated?
Each model is scored per category from its published benchmark results, then the categories are weighted into one figure: agentic 22%, coding 20%, reasoning 17%, multimodal 12%, knowledge 12%, multilingual 7%, instruction following 5%, maths 5%. WriteWorks holds scores for 35 of 418 tracked models across 19 benchmarks. A category with no published evidence is left out and the rest re-weighted, never counted as zero.
Which LLM is best?
There is no single best model, and any leaderboard implying otherwise is answering a narrower question than it looks. The top of a general ranking is the model with the strongest average across weighted categories; the right model for a given job is usually the one leading the one category that job depends on, at a price the workload can carry.
How often are the rankings and benchmarks updated?
Model catalogues are re-read every three hours and provider pricing pages every six, so releases and price moves appear the same day. Benchmark scores are added as results are published, and every row on a model page carries the date it was confirmed.
What is the difference between LLM benchmarks and an LLM leaderboard?
A benchmark is one test with one score, such as SWE-bench for resolving real code issues. A leaderboard is the ranking that falls out of combining several and deciding what each is worth. The benchmarks are the evidence; the leaderboard is an argument about how to weigh it, which is why this one publishes its weights.
Are these AI benchmarks independently verified?
Each score is labelled. Verified means the score was read on the benchmark's own published results. Provider claim means the only source is the company that made the model. Both are shown, because dropping everything that cannot be verified independently would quietly favour the labs that publish least.
WriteWorks

Are these models citing you, or your competitors?

WriteWorks tracks how often 10+ AI platforms, from ChatGPT and Claude to Gemini and Perplexity, mention your brand against your competitors, then helps you engineer the content that changes the answer.