Skip to content
Which AI

Models

Which AI model is best right now?

Every model from the five providers we track, ranked on independent benchmarks — not the figures labs choose to publish at launch. Next to each, the cheapest subscription that includes it.

Ranked by benchmarks

Latest result 29 Sept 2026

General and Coding are our 0–100 scores, the same ones that decide a plan's model quality in the rankings. Click a column to sort by it. A dash means too few relevant benchmarks to score fairly.

AI models ranked by independent benchmarks
#Model
1
Claude Opus 5.5
In Claude Pro · $20/mo and 2 more
89—
2
Claude Fable 5.1
Not in a tracked subscription
8079
3
GPT-6 Astra
In ChatGPT Pro · $100/mo and 2 more
7978
4
Claude Opus 5
In Claude Pro · $20/mo and 2 more
7474
5
Claude Fable 5
Not in a tracked subscription
7271
6
GPT-5.5 Pro
Not in a tracked subscription
72—
7
Claude Sonnet 5.5
In Claude Pro · $20/mo and 2 more
72—
8
GPT-5.4 Pro
Not in a tracked subscription
65—
9
Claude Opus 4.8
Not in a tracked subscription
63—
10
GPT-5.6 Sol
In ChatGPT Plus · $20/mo and 2 more
6364
11
GPT-5.5
Not in a tracked subscription
6161
12
Grok 4.6
In SuperGrok · $30/mo and 2 more
6061
13
GPT-5.4
Not in a tracked subscription
60—
14
Gemini 3.7 Flash
In Google AI Pro · $19.99/mo and 2 more
6059
15
Gemini 3.8 Flash
In Google AI Pro · $19.99/mo and 4 more
5959
16
GPT-5.6 Luna
Not in a tracked subscription
5757
17
GPT-5.6 Terra
Not in a tracked subscription
5757
18
GPT-5.2 Pro
Not in a tracked subscription
56—
19
Claude Sonnet 5
In Claude Pro · $20/mo and 4 more
5655
20
GPT-6 Sol
In ChatGPT Plus · $20/mo and 4 more
5455
21
Claude Opus 4.7
Not in a tracked subscription
5353
22
GPT-5.3-Codex
Not in a tracked subscription
52—
23
GPT-5.2
Not in a tracked subscription
52—
24
Gemini 3.5 Flash
Not in a tracked subscription
5150
25
Grok 4.5
Not in a tracked subscription
5150
26
GPT-6 Luna
Not in a tracked subscription
5049
27
Grok 4.20
Not in a tracked subscription
49—
28
Claude Opus 4.6
Not in a tracked subscription
4848
29
Gemini 3.6 Flash
In Google AI Plus · $4.99/mo
4742
30
GPT-5 Pro
Not in a tracked subscription
45—
31
Claude Opus 4.5
Not in a tracked subscription
4545
32
GPT-5
Not in a tracked subscription
44—
33
Claude Sonnet 4.6
Not in a tracked subscription
4444
34
GPT-5.1
Not in a tracked subscription
44—
35
Gemini 3 Flash Preview
Not in a tracked subscription
43—
36
Gemini 3.1 Pro Preview
In Google AI Pro · $19.99/mo and 3 more
4129
37
o3 Pro
Not in a tracked subscription
39—
38
GPT-5.2-Codex
Not in a tracked subscription
38—
39
Claude Sonnet 4.5
Not in a tracked subscription
37—
40
GPT-5.4 Mini
Not in a tracked subscription
3636
41
Gemini 3.5 Flash Lite
Not in a tracked subscription
3637
42
GPT-5.4 Nano
Not in a tracked subscription
35—
43
o4 Mini
Not in a tracked subscription
35—
44
o3
Not in a tracked subscription
3434
45
GPT-5.1-Codex
Not in a tracked subscription
32—
46
Claude Opus 4.1
Not in a tracked subscription
31—
47
Claude Haiku 4.5
Not in a tracked subscription
28—
48
GPT-5 Mini
Not in a tracked subscription
2724
49
Gemini 3.1 Flash Lite
Not in a tracked subscription
27—
50
o1
Not in a tracked subscription
26—
51
Claude Sonnet 4
Not in a tracked subscription
26—
52
Gemma 4 31B
Not in a tracked subscription
26—
53
Gemini 2.5 Pro
Not in a tracked subscription
2423
54
Gemini 2.5 Pro Preview 06-05
In Google AI Pro · $19.99/mo and 2 more
22—
55
gpt-oss-120b
Not in a tracked subscription
22—
56
Gemini 2.5 Flash
Not in a tracked subscription
20—
57
o3 Mini
Not in a tracked subscription
2019
58
GPT-4.1
Not in a tracked subscription
1717
59
gpt-oss-20b
Not in a tracked subscription
1718
60
GPT-5 Nano
Not in a tracked subscription
1616
61
GPT-4.1 Mini
Not in a tracked subscription
11—
62
Gemini 2.5 Flash Lite
Not in a tracked subscription
10—
63
GPT-3.5 Turbo
Not in a tracked subscription
99
64
GPT-4.1 Nano
Not in a tracked subscription
0—
65
GPT-4o-mini
Not in a tracked subscription
0—

Not enough benchmarks yet

Measured on one or two narrow benchmarks so far — usually a new release that Epoch AI and Artificial Analysis have not finished testing. Listed, not ranked, until they have.

  • GPT-5.1-Codex-Mini
  • Grok 4.3
  • Grok 4.7

How the scores work

Each benchmark is placed on a fixed 0–100 scale, then combined by how much it says about general or coding ability. A model is only scored on its own results; one measured on too little is not ranked. Where a model was tested at several effort settings, its best counts. The full method is on the methodology page.

Results from Epoch AI (Epoch Capabilities Index, DeepSWE, FrontierCode, CursorBench, WebDev Arena; data © Epoch AI, CC BY 4.0) and Artificial Analysis (Intelligence and Coding indexes), refreshed nightly.