AI Benchmarks

AI model leaderboard

How the leading AI models compare on quality, coding, speed and cost. Use the buttons to sort.

Frontier model ranking

Source: Artificial Analysis Intelligence Index v5.0.1 · updated 16 Nov 2026

#ModelIndexArtificial Analysis Intelligence Index: an overall 0–100 score from dozens of tests (reasoning, knowledge, maths, coding, tool use). Higher means a more capable model.CodingTerminal-Bench 4.0 (measured by Artificial Analysis): the model works like a developer inside a computer terminal and has to finish real tasks on its own, such as fixing code or setting up a program. The percentage shows how many tasks it completed.SpeedHow many tokens (pieces of words) the model writes per second through the API, median value. A dash means it has not been measured yet.Price / 1M tokensAPI cost in US dollars per 1 million tokens: “in” for what you send to the model, “out” for the answer it writes. 1M tokens is roughly 750,000 English words.
1 Claude Fable 5.2 (max) Anthropic Index65 60% 102 tok/s $40.00 out$8.00 in
2 Gemini 4 Carbon (high) Google DeepMind Index60 65% 79 tok/s $20.00 out$4.00 in
3 Claude Opus 5.5 (max) Anthropic Index59 60% 115 tok/s $20.00 out$4.00 in
4 GPT-6 Astra (max) OpenAI Index57 59% 37 tok/s $50.00 out$10.00 in
5 Claude Sonnet 5.5 (max) Anthropic Index55 64% 151 tok/s $10.00 out$2.00 in
6 GPT-6.1 Sol (max) OpenAI Index55 56% 57 tok/s $10.00 out$2.00 in
7 Gemini 4 Hydrogen (xhigh) Google DeepMind Index54 54% – $7.50 out$1.50 in
8 GPT-6.1 Terra (max) OpenAI Index51 56% 74 tok/s $5.00 out$1.00 in
9 Gemini 4 Argon (high) Google DeepMind Index50 57% 67 tok/s $20.00 out$4.00 in
10 Muse Wonder Full (max) Meta AI Index50 57% – –
11 Kimi K3.1 Pro Open weights Moonshot AI Index47 67% 32 tok/s $10.00 out$2.00 in
12 Muse Wonder (max) Meta AI Index44 52% 125 tok/s $20.00 out$4.00 in
13 Claude Haiku 5.5 (max) Anthropic Index37 33% 254 tok/s $0.50 out$0.10 in
14 Muse Ember (max) Meta AI Index36 41% 446 tok/s $0.50 out$0.10 in
15 Grok 4.7 (high) xAI Index35 25% 69 tok/s $6.00 out$2.00 in
16 Qwen3.8 Max (0902) Alibaba Index34 39% 50 tok/s $6.00 out$2.00 in
17 DeepSeek V4.1 Flash 1017 (max) Open weights DeepSeek Index33 34% 140 tok/s $1.20 out$0.30 in
18 GLM-5.3 (max) Open weights Zhipu AI Index32 42% 66 tok/s $4.40 out$1.40 in
19 Gemini 3.8 Flash (high) Google DeepMind Index31 20% 112 tok/s $3.75 out$0.75 in
20 Ling 3.1 Flash 1019 Open weights InclusionAI Index30 33% – $0.90 out$0.30 in
21 MiMo-V2.6-Pro Open weights Xiaomi Index30 35% 55 tok/s $0.87 out$0.43 in
22 MiniMax-M3.1-Flash Open weights MiniMax Index29 26% 160 tok/s $0.60 out$0.15 in
23 GPT-6 Luna (max) OpenAI Index28 13% 104 tok/s $0.50 out$0.10 in
24 Mistral Large 4 (high) Open weights Mistral AI Index26 25% 109 tok/s $2.10 out$0.70 in
What do the columns mean?
Index
Artificial Analysis Intelligence Index: an overall 0–100 score from dozens of tests (reasoning, knowledge, maths, coding, tool use). Higher means a more capable model.
Coding
Terminal-Bench 4.0 (measured by Artificial Analysis): the model works like a developer inside a computer terminal and has to finish real tasks on its own, such as fixing code or setting up a program. The percentage shows how many tasks it completed.
Speed
How many tokens (pieces of words) the model writes per second through the API, median value. A dash means it has not been measured yet.
Price
API cost in US dollars per 1 million tokens: “in” for what you send to the model, “out” for the answer it writes. 1M tokens is roughly 750,000 English words.
(max), (xhigh)…
The level of “thinking” (reasoning effort) the model was tested with. More thinking usually gives better results, but slower and at a higher cost.
Open weights
The model’s “weights” are publicly available, so it can be installed and run on your own computers.
Index: Artificial Analysis Intelligence Index (0–100) · Coding: Terminal-Bench 4.0 · Prices in $ per 1M tokens (input / output)