Independent AI model guide · Updated September 25, 2026
Know which model is worth it, and what it costs to run.
267 language, image, and video models from 66 labs, ranked on independent benchmarks and priced from the source. Then a planner that tells you whether to rent tokens or buy the tower.
- Models tracked
- 267
- Benchmarks explained
- 32
- Sourced scores
- 2,010
- Labs covered
- 66
The frontier
Capability is getting cheaper. Not every price has noticed.
Each dot is a model: how it scores against what a million tokens costs. The line connects the models that nothing cheaper beats. MiMo-V2.6-Pro keeps 80% of Claude Opus 5.5’s score at $0.57 per million tokens, 16× cheaper.
Intelligence Index against blended price
92 language models with an independent index score and a list price. Blended price is 70% input, 30% output.
- Closed weights
- Open weights
- Efficient frontier
View as table
| Model | Provider | AA Intelligence Index | Blended $/M | Frontier |
|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic | 57.6 | $8.80 | Yes |
| Claude Sonnet 5.5 | Anthropic | 56 | $4.40 | Yes |
| Claude Fable 5.1 | Anthropic | 53.4 | $22.00 | |
| Gemini 4 Argon | 53 | $4.40 | ||
| GPT-6 Astra | OpenAI | 52.7 | $22.00 | |
| GPT-6.1 Sol | OpenAI | 52 | $4.40 | |
| Muse Wonder | Meta | 52 | $8.80 | |
| Claude Opus 5 | Anthropic | 50.8 | $11.00 | |
| Claude Fable 5 | Anthropic | 49.6 | $22.00 | |
| Muse Spark 1.3 | Meta | 48.1 | $2.15 | Yes |
| GPT-6 Sol | OpenAI | 47.5 | $4.40 | |
| GPT-5.6 Sol | OpenAI | 47 | $8.80 | |
| Grok 4.7 | xAI | 46.4 | $3.20 | |
| MiMo-V2.6-Pro | Xiaomi | 46.3 | $0.57 | Yes |
| Qwen3.8-Max | Alibaba | 45 | $3.20 | |
| GLM-5.3 | Z.ai | 45 | $2.30 | |
| Grok 4.6 | xAI | 44.3 | $3.20 | |
| Kimi K3 | Moonshot AI | 44 | $6.60 | |
| Step 5 Preview | StepFun | 43.7 | $1.51 | |
| Claude Haiku 5.5 | Anthropic | 43 | $0.22 | Yes |
| GPT-5.6 Terra | OpenAI | 42.1 | $5.00 | |
| GLM-5.3-Flash | Z.ai | 42 | $0.26 | |
| Claude Opus 4.8 | Anthropic | 41.8 | $11.00 | |
| Gemini 3.8 Flash | 40.9 | $1.65 | ||
| Qwen3.8-2.4T-A95B | Alibaba | 40 | $3.20 | |
| Muse Spark 1.2 | Meta | 39.6 | $2.15 | |
| Gemini 3.7 Flash | 39.1 | $1.65 | ||
| DeepSeek V4.1 Flash | DeepSeek | 39 | $0.57 | |
| Grok 4.5 | xAI | 38.8 | $3.20 | |
| Claude Sonnet 5 | Anthropic | 38.2 | $4.40 | |
| Mistral Large 4 | Mistral | 38 | $2.21 | |
| GPT-6 Luna | OpenAI | 37.3 | $0.22 | |
| GPT-5.6 Luna | OpenAI | 37.3 | $0.50 | |
| DeepSeek V4 Pro (0813) | DeepSeek | 36 | $2.11 | |
| Gemini 3.6 Flash | 34 | $1.65 | ||
| Qwen3.8-27B | Alibaba | 34 | $1.25 | |
| GLM-5.2 | Z.ai | 34 | $2.30 | |
| Muse Spark 1.1 | Meta | 33.7 | $2.15 | |
| Gemini 3.5 Flash | 32.6 | $3.75 | ||
| GPT-5.3 Codex | OpenAI | 32.5 | $5.43 | |
| Gemini 3.1 Pro Preview | 29.7 | $5.00 | ||
| Qwen3.7-Max | Alibaba | 29 | $4.00 | |
| MiniMax-M3 | MiniMax | 29 | $0.57 | |
| GLM-5 | Z.ai | 28 | $1.66 | |
| Inkling-Small | Thinking Machines Lab | 27.8 | $0.57 | |
| Grok Build 0.1 | xAI | 27.2 | $1.30 | |
| Kimi K2.6 | Moonshot AI | 27 | $1.87 | |
| Quasar 438B | Multiverse Computing | 26.7 | $0.96 | |
| Apodex 1.1 | Apodex | 26.4 | $1.11 | |
| Kimi K2.7 Code | Moonshot AI | 26 | $1.87 | |
| MiMo-V2.5-Pro | Xiaomi | 26 | $0.57 | |
| Tencent Hy3 | Tencent | 25.3 | $0.16 | Yes |
| MiMo-V2.5 | Xiaomi | 25.2 | $0.18 | |
| Qwen3.7-Plus | Alibaba | 25 | $0.76 | |
| Inkling | Thinking Machines Lab | 25 | $1.92 | |
| Grok 4.3 | xAI | 24.9 | $1.63 | |
| GPT-5.4 mini | OpenAI | 24.1 | $1.88 | |
| Solar Mini 4 | Upstage | 24 | $0.19 | |
| MiniMax-M2.7 | MiniMax | 23 | $0.57 | |
| Nemotron 3 Ultra 550B A55B | NVIDIA | 22.9 | $1.17 | |
| Gemini 3.5 Flash-Lite | 22.2 | $0.96 | ||
| Qwen3.6-27B | Alibaba | 21 | $1.50 | |
| GPT-5.4 nano | OpenAI | 20.7 | $0.52 | |
| Step 3.7 Flash | StepFun | 19.5 | $0.49 | |
| LongCat-2.0 | Meituan | 19.1 | $1.41 | |
| Qwen3.6-35B-A3B | Alibaba | 18 | $0.94 | |
| Claude Haiku 4.5 | Anthropic | 16.9 | $2.20 | |
| Granite 4.2 30B | IBM | 14.8 | $0.31 | |
| Mistral Medium 3.5 | Mistral | 14.2 | $3.30 | |
| Amazon Nova 2 Pro (Preview) | Amazon | 14.2 | $3.88 | |
| Amazon Nova 2 Omni (Preview) | Amazon | 13.6 | $0.96 | |
| Amazon Nova 2 Lite | Amazon | 13.4 | $0.96 | |
| Nemotron 3.5 Lightning 30B A3B | NVIDIA | 12.9 | $0.12 | Yes |
| Nemotron 3 Super 120B A12B | NVIDIA | 12.8 | $0.48 | |
| Mercury 2.5 | Inception | 12.3 | $0.40 | |
| Mistral Small 4 | Mistral | 11.3 | $0.29 | |
| Granite 4.2 8B | IBM | 11.1 | $0.12 | |
| Trinity Large Thinking | Arcee AI | 10.8 | $0.42 | |
| Nemotron 3 Nano Omni 30B A3B Reasoning | NVIDIA | 10.3 | $0.47 | |
| Mistral Large 3 | Mistral | 9.3 | $0.80 | |
| Granite 4.2 3B | IBM | 9.1 | $0.06 | Yes |
| Nemotron 3 Nano 30B A3B | NVIDIA | 8.9 | $0.10 | |
| Hermes 4 405B | Nous Research | 7.5 | $1.60 | |
| Amazon Nova Pro | Amazon | 7 | $1.52 | |
| Command A | Cohere | 7 | $4.75 | |
| Amazon Nova Lite | Amazon | 6.7 | $0.11 | |
| Ministral 3 14B | Mistral | 6 | $0.20 | |
| Amazon Nova Micro | Amazon | 5.9 | $0.07 | |
| Reka Flash 3 | Reka AI | 5.6 | $0.38 | |
| Ministral 3 8B | Mistral | 5.5 | $0.15 | |
| Apertus 70B Instruct | Swiss AI Initiative | 5.1 | $1.45 | |
| Ministral 3 3B | Mistral | 4.8 | $0.10 |
Who leads what
There is no best model. There is a best model for the job.
The overall index hides a lot. These are the leaders on the tests closest to real work, with the runner-up so you can see how close the race is.
Fixing real repositories
Claude Opus 5.589.9%vendor
SWE-bench Pro · next: Claude Sonnet 5.5 81.3%
Working in a terminal
Claude Sonnet 5.570.6%vendor
Terminal-Bench 4.0 · next: Claude Mythos 5.1 60.9%
Following policy with tools
GLM-5.299.1%
τ²-Bench · next: Claude Fable 5 98.5%
Expert-level questions
Claude Sonnet 5.564.5%vendor
Humanity’s Last Exam · next: Claude Opus 5.5 61.4%
Generating video
Wan 3.01156
Artificial Analysis Video Arena: text to video with audio · next: Utopai X 1149
Generating images
GPT Image 2.5 Sunburst1196
Artificial Analysis Image Arena: text to image · next: GPT Image 2.5 Flare 1190
Just released
The newest models, already on the board.
Oct 12Muse WonderMetaIndex52.0In / out per 1M$4.00 / $20.00- Oct 7Claude Haiku 5.5AnthropicIndex43.0In / out per 1M$0.10 / $0.50
Oct 7Vidu Q4 PreviewShengShuIndex—Price$0.08 per second of 1080p video, as listed by Artificial Analysis ($5.04/min); Vidu has not published Q4 on its pricing page
Oct 6Mistral Large 4MistralIndex38.0In / out per 1M$1.36 / $4.18
Oct 6Nano Banana 2.1GoogleIndex—In / out per 1M$1.50 / $30.00
Oct 5BeamReflection AIIndex—In / out per 1MNo list price
Oct 3Kolibri-1Aleph AlphaopenIndex—In / out per 1MSelf-host
Oct 1Grok Imagine Video 1.5 LitexAIIndex—Price$0.02 per second of video (xAI lists one rate; Artificial Analysis lists $8.40/min at 1080p)
Four ways in
Compare, plan, build, then check our work.
01
Compare
02
Plan
03
Build
04
Lab
From the lab
We test the claims ourselves.
Stack planner
Six questions. Three priced stacks.
Tell us the work, the data rules, and the budget. The planner weighs every API price against hardware you could own and shows the math behind each pick.
Plan a stack