Updated Sep 22, 2026

AI Models Leaderboard, September 2026

Large language models ranked by LMSys Arena Elo, MMLU, HumanEval, MATH, pricing, and inference speed. Refreshed monthly with live data from official provider pricing pages, Artificial Analysis, and the Arena.

In short:As of September 2026, Claude Mythos 5 tops the AI models leaderboard at 1531 Arena Elo, across 56+ models compared on quality, cost, and speed. Per-task leaders differ: check the column for your use case (text, coding, reasoning, or value).

Which AI model should you use in September 2026?

The leaderboard below ranks 56 frontier and frontier-adjacent models on a single normalized quality index, but the practical answer is task-dependent: pick the leading code/agentic model for coding, the leading long-context model for science and retrieval, the leading chat model for general use, and the open-weights leader when budget matters. Multi-model routing is now the default architectural choice for serious production deployments.

56 models
#ModelQualityArena ELOSpeedPriceContextValueReleased
1

Anthropic · Frontier agentic coding & knowledge work

100
152558 t/s$10 / $501M3.3Jun 2026
2

Anthropic · Ceiling capability (limited access)

100
153156 t/s$10 / $501M3.3Jun 2026
3

Anthropic · Frontier agentic coding & reasoning

99
152274 t/s$5 / $251M6.6Jul 2026
4

Anthropic · Coding, agents & computer use

98
151272 t/s$5 / $251M6.5May 2026
5

OpenAI · General frontier reasoning & tools

98
151496 t/s$5 / $301M5.6Jul 2026
6

OpenAI · Frontier general purpose

97
150670 t/s$5 / $301M5.5Apr 2026
7

Moonshot AI · Open-weight coding frontier

97
150055 t/s$3 / $151M10.8Jul 2026
8

OpenAI · Reasoning at any cost

96
151068 t/s$30 / $1801M0.9Apr 2026
9

Anthropic · Coding & agentic workflows

96
150568 t/s$5 / $251M6.4Apr 2026
10

Google · Science & long-context

96
1505131 t/s$2 / $121M13.7Apr 2026
11

Alibaba Cloud · Multimodal APAC frontier (hosted); text-only open weights

96
149186 t/s$2 / $61M24.0Aug 2026
12

OpenAI · Hard reasoning

94
137068 t/s$2 / $8200K18.8Apr 2025
13

Alibaba Cloud · Long autonomous agentic runs

94
148890 t/s$2.5 / $7.51M18.8May 2026
14

xAI · Cheap output tokens & real-time info

94
149988 t/s$2 / $6500K23.5Jul 2026
15

xAI · Agentic tasks & real-time info

93
149683 t/s$1.25 / $2.51M49.6May 2026
16

OpenAI · Default production tier

93
1489118 t/s$2 / $121M13.3Jul 2026
17

Anthropic · Balanced agents & coding value

93
147998 t/s$3 / $151M10.3Jun 2026
18

Google · Multimodal + value

92
134587 t/s$1.25 / $101M16.4Mar 2025
19

Moonshot AI · Frontier quality at low cost

92
146648 t/s$0.73 / $3.49256K43.6Apr 2026
20

Anthropic · Complex analysis

91
136052 t/s$15 / $75200K2.0May 2025
21

DeepSeek · Cheap reasoning

91
135035 t/s$0.55 / $2.19128K66.4Jan 2025
22

Z.ai (Zhipu AI) · Fastest self-hostable frontier model

91
1483168 t/s$1.4 / $4.41M31.4Jun 2026
23

Anthropic · Coding & balance

90
146773 t/s$3 / $151M10.0Feb 2026
24

Google · Fast multimodal at mid-tier price

90
1481148 t/s$0.75 / $3.751M40.0Jul 2026
25

OpenAI · Long context

89
1310120 t/s$2 / $81M17.8Apr 2025
26

DeepSeek · Cheap frontier-adjacent tokens (GA build)

89
146162 t/s$0.66 / $1.981M67.4Aug 2026
27

MiniMax · Open-weight agentic coding

89
145580 t/s$0.6 / $2.41M59.3Jun 2026
28

OpenAI · Reasoning & math

88
1305155 t/s$1.1 / $4.4200K32.0Jan 2025
29

Anthropic · Coding & balance

88
132095 t/s$3 / $15200K9.8May 2025
30

Z.ai (Zhipu AI) · Open-weight agentic & tool use

88
146748 t/s$0.98 / $3.08200K43.3Apr 2026
31

xAI · Real-time info

87
133082 t/s$3 / $15131K9.7Feb 2025
32

DeepSeek · Best open-source value

86
131062 t/s$0.27 / $1.1128K125.5Mar 2025
33

Alibaba Cloud · Multilingual & APAC

86
1448124 t/s$1.4 / $5.6256K24.6Apr 2026
34

OpenAI · General purpose

85
1285109 t/s$2.5 / $10128K13.6May 2024
35

DeepSeek · Cheapest tokens, period

84
1441105 t/s$0.14 / $0.281M400.0Apr 2026
36

OpenAI · High-volume cheap frontier-family

83
1436186 t/s$0.2 / $1.21M118.6Jul 2026
37

Meta · Open-source value

80
1260135 t/s$0.2 / $0.61M200.0Apr 2025
38

Alibaba Cloud · Open-source flagship

80
125585 t/s$0.3 / $0.9131K133.3Sep 2024
39

Alibaba Cloud · Locally-runnable open-weight (~17GB VRAM at 4-bit)

80
Self-host262K, Aug 2026
40

Mistral AI · Multilingual

79
125078 t/s$2 / $6128K19.8Nov 2024
41

Google · High-volume cheap multimodal

79
1418192 t/s$0.3 / $2.51M56.4Jul 2026
42

NVIDIA · Efficient single-GPU deployment, 1M context

79
Self-host1M, Aug 2026
43

xAI · Budget reasoning

78
1275165 t/s$0.3 / $0.5131K195.0Feb 2025
44

Perplexity · Search + citations

78
65 t/s$3 / $15200K8.7Feb 2025
45

Mistral AI · Code generation

76
195 t/s$0.3 / $0.9256K126.7Jan 2025
46

NVIDIA · Open multimodal

76
1361158 t/sSelf-host256K, Apr 2026
47

Tencent · Long-context retrieval & agentic search, no license restrictions

76
Self-host256K, Jul 2026
48

Anthropic · Speed & cost

75
1230172 t/s$0.8 / $4200K31.3Oct 2024
49

Google · Self-hosted general purpose

75
1351142 t/sSelf-host128K, Apr 2026
50

Google · Fastest + cheapest

74
1240244 t/s$0.1 / $0.41M296.0Feb 2025
51

Alibaba Cloud · Open-source coding

74
125 t/s$0.15 / $0.45131K246.7Nov 2024
52

Ant Group · Efficient MoE alternative to trillion-param flagships

74
Self-host256K, Aug 2026
53

OpenAI · High throughput

72
1216183 t/s$0.15 / $0.6128K192.0Jul 2024
54

Meta · Longest context

71
1195198 t/s$0.15 / $0.410M258.2Apr 2025
55

Amazon · AWS ecosystem

70
110 t/s$0.8 / $3.2300K35.0Dec 2024
56

Cohere · Enterprise RAG

68
117072 t/s$2.5 / $10128K10.9Aug 2024
Quality = composite benchmark (MMLU, HumanEval, MATH)Arena ELO = LMSYS Chatbot Arena ratingValue = quality per dollarPrice = input / output per 1M tokens

How the LLM leaderboard works

We pull official provider pricing every 24 hours, Artificial Analysis benchmark snapshots weekly, and LMSys Arena Elo as it publishes. The composite quality index is a 0-100 normalization over MMLU Pro, HumanEval, and MATH, weighted by recency and cross-validated against Arena Elo. We do not accept vendor-supplied numbers without an independent reference.

Where the leaderboard is wrong

No leaderboard predicts your production accuracy. LMSys Arena rewards style and short-conversation polish; a top-Arena model can still under-perform on your specific function-calling schema or long-context retrieval workload. Build an internal eval harness before you commit. See our LMArena Elo explained and LLM routing writeups for the deep-dive.

Related rankings

Deploy a model with Swfte Connect

One gateway, every provider, per-token cost visibility. Swap models without touching your code.