LLM Wrapped 2026
265
AI models released in 2026 so far.
open weights (137) closed (128)One square per release, 1 January first.
Through 31 Oct 2026; this page updates every day until the year ends.
Apr was the busiest month
42 releases, out of 265 this year.
Google shipped the most
32 models from one lab. The six busiest:
Google32
Alibaba28
OpenAI24
NVIDIA15
Anthropic13
Zhipu AI10
The top spot changed hands 10 times
Each model that, on its release day, scored above everything released before it. From 61.0 on the BenchLeader Index to 69.2: 8.2 points in the year so far.
- On 1 Jano361.0
- 5 FebClaude Opus 4.662.6
- 5 FebGPT-5.3-Codex62.9
- 4 MarGPT-5.4 Pro63.1
- 5 MarGPT-5.463.3
- 8 Apr
Muse Spark64.3 - 23 AprGPT-5.564.4
- 9 JunClaude Fable 566.6
- 1 SeptClaude Fable 5.167.1
- 2 SeptGPT-6 Astra68.2
- 22 SeptClaude Opus 5.569.2
The year's standouts
Among models released in 2026.
- Model of the yearClaude Opus 5.5Index 69.2Released 22 Sept 2026
- Best at codingGPT-6 AstraCoding 83Coding category of the Index
- Fastest good model
Grok 4.20 Multi-Agent360 tok/sIndex above 55 - Longest thinkerClaude Sonnet 5.5285k tokensOutput per Intelligence Index task
The same quality got 2.3× cheaper
What the best model on 1 January scored (61.0), and the cheapest way to get it then and now, at today's list prices.
Open weights fell further behind: 2.2 points to 7.0
Best Index among models released by each month's end, open weights against everything else.
Day by day at #1
Since BenchLeader began tracking daily on 8 Sept 2026: who led the Index each day.
- GPT-6 Astra21 days
- Claude Opus 5.533 days
Share it
How this is counted: releases are every model in BenchLeader’s registry with a release date in 2026, serving tiers aside. Rankings use each model’s best reasoning setting and today’s scores, not the ones it had on launch day, so a frontier “on 1 March” is the best model released by then as measured now. Prices are today’s list prices. How the Index works.