Start free
Live · The State of AI

What AI actually costs, right now.

One live, citable view of the whole picture: the AI Cost Index, what our community really spends, the biggest price moves, and every model price you can filter and sort yourself. Built from real usage. Free under CC BY 4.0.

0

models priced

0

providers tracked

$0

community spend tracked

0

tokens tracked

Prices updated 9 hours ago · refreshes automatically

fireworks_ai/accounts/fireworks/models/flux-1-dev-controlnet-union $0 /Mtok mistral/mistral-small-latest $0.09 /Mtok nebius/Qwen/Qwen3-14B $0.12 /Mtok openrouter/mistralai/mistral-small-3.2-24b-instruct $0.15 /Mtok fireworks_ai/accounts/fireworks/models/llama-v3p2-11b-vision-instruct $0.2 /Mtok fireworks_ai/accounts/fireworks/models/rolm-ocr $0.2 /Mtok fireworks_ai/accounts/fireworks/models/qwen3-vl-30b-a3b-instruct $0.26 /Mtok watsonx/meta-llama/llama-3-2-11b-vision-instruct $0.35 /Mtok azure/gpt-5.4-nano $0.46 /Mtok azure_ai/jamba-instruct $0.55 /Mtok openrouter/openai/gpt-5-mini $0.69 /Mtok openrouter/qwen/qwen3.5-122b-a10b $0.8 /Mtok fireworks_ai/accounts/fireworks/models/deepseek-coder-33b-instruct $0.9 /Mtok together_ai/zai-org/GLM-4.6 $1 /Mtok fireworks_ai/accounts/fireworks/models/deepseek-prover-v2 $1.2 /Mtok ft:babbage-002 $1.6 /Mtok watsonx/meta-llama/llama-3-2-90b-vision-instruct $2 /Mtok mistral/mistral-medium-2604 $3 /Mtok replicate/openai/gpt-4.1 $3.5 /Mtok vertex_ai/gemini-3.1-pro-preview $4.5 /Mtok gmi/anthropic/claude-sonnet-4 $6 /Mtok anthropic.claude-3-7-sonnet-20240620-v1:0 $7.2 /Mtok jp.anthropic.claude-opus-4-8 $11 /Mtok claude-opus-4-1-20250805 $30 /Mtok fireworks_ai/accounts/fireworks/models/flux-1-dev-controlnet-union $0 /Mtok mistral/mistral-small-latest $0.09 /Mtok nebius/Qwen/Qwen3-14B $0.12 /Mtok openrouter/mistralai/mistral-small-3.2-24b-instruct $0.15 /Mtok fireworks_ai/accounts/fireworks/models/llama-v3p2-11b-vision-instruct $0.2 /Mtok fireworks_ai/accounts/fireworks/models/rolm-ocr $0.2 /Mtok fireworks_ai/accounts/fireworks/models/qwen3-vl-30b-a3b-instruct $0.26 /Mtok watsonx/meta-llama/llama-3-2-11b-vision-instruct $0.35 /Mtok azure/gpt-5.4-nano $0.46 /Mtok azure_ai/jamba-instruct $0.55 /Mtok openrouter/openai/gpt-5-mini $0.69 /Mtok openrouter/qwen/qwen3.5-122b-a10b $0.8 /Mtok fireworks_ai/accounts/fireworks/models/deepseek-coder-33b-instruct $0.9 /Mtok together_ai/zai-org/GLM-4.6 $1 /Mtok fireworks_ai/accounts/fireworks/models/deepseek-prover-v2 $1.2 /Mtok ft:babbage-002 $1.6 /Mtok watsonx/meta-llama/llama-3-2-90b-vision-instruct $2 /Mtok mistral/mistral-medium-2604 $3 /Mtok replicate/openai/gpt-4.1 $3.5 /Mtok vertex_ai/gemini-3.1-pro-preview $4.5 /Mtok gmi/anthropic/claude-sonnet-4 $6 /Mtok anthropic.claude-3-7-sonnet-20240620-v1:0 $7.2 /Mtok jp.anthropic.claude-opus-4-8 $11 /Mtok claude-opus-4-1-20250805 $30 /Mtok

Key findings

What the numbers say right now

85% quality, 5% price

GPT-5.6 Luna (max) reaches 85% of the top intelligence score at 5% of the price

Versus Claude Opus 5 (Adaptive Reasoning, Xhigh Effort), the current intelligence leader. Quality from Artificial Analysis; price blended 3:1.

61,538 pages for $1

One dollar now buys about 61,538 pages of AI-written text

At Gemma 3n E4B Instruct, one of the cheapest capable models, $0.03 per million tokens (blended).

$3.25

A 100,000-word novel costs about $3.25 to generate on Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

Roughly 130,000 output tokens on today's top-rated model.

10,500x spread

The priciest model costs about 10,500x more than the cheapest for the same million tokens

o1-pro vs Gemma 3n E4B Instruct, blended price. Picking the right model is the biggest cost lever you have.

$0.06 / 1M

Open-weight models now run as low as $0.06 per million tokens

Qwen3.5 4B (Non-reasoning) is the cheapest open-weight model tracked, roughly 25,641 pages per dollar.

Computed live from our price, quality, and index data. Free to use and cite under CC BY 4.0.

Benchmark

The AI Cost Index

What a million tokens costs across a fixed basket of models, tracked over time. The S&P 500 of AI prices.

Frontier

One flagship model per major provider

$4.64/Mtok

0% since start

$3.40 $4.23 $5.05 $5.88 Jun 15: $4.64 Jun 16: $4.64 Jun 17: $4.64 Jun 19: $4.64 Jun 20: $4.64 Jun 21: $4.64 Jun 22: $4.64 Jun 23: $4.64 Jun 24: $4.64 Jun 25: $4.64 Jun 26: $4.64 Jun 27: $4.64 Jun 28: $4.64 Jun 29: $4.64 Jun 30: $4.64 Jul 1: $4.64 Jul 2: $4.64 Jul 3: $4.64 Jul 4: $4.64 Jul 5: $4.64 Jul 6: $4.64 Jul 7: $4.64 Jul 8: $4.64 Jul 9: $4.64 Jul 10: $4.64 Jul 11: $4.64 Jul 12: $4.64 Jul 13: $4.64 Jul 14: $4.64 Jul 15: $4.64 Jul 16: $4.64 Jul 17: $4.64 Jul 18: $4.64 Jul 19: $4.64 Jul 20: $4.64 Jul 21: $4.64 Jul 22: $4.64 Jul 23: $4.64 Jul 24: $4.64 Jul 25: $4.64 Jul 26: $4.64 Jul 27: $4.64 Jul 28: $4.64 Jul 29: $4.64 Jul 30: $4.64 Jul 31: $4.64 Jun 15 Jul 31

Budget

One cost-efficient workhorse per major provider

$0.7/Mtok

0% since start

$0.00 $0.64 $1.27 $1.91 Jun 15: $0.70 Jun 16: $0.70 Jun 17: $0.70 Jun 19: $0.70 Jun 20: $0.70 Jun 21: $0.70 Jun 22: $0.70 Jun 23: $0.70 Jun 24: $0.70 Jun 25: $0.70 Jun 26: $0.70 Jun 27: $0.70 Jun 28: $0.70 Jun 29: $0.70 Jun 30: $0.70 Jul 1: $0.70 Jul 2: $0.70 Jul 3: $0.70 Jul 4: $0.70 Jul 5: $0.70 Jul 6: $0.70 Jul 7: $0.70 Jul 8: $0.70 Jul 9: $0.70 Jul 10: $0.70 Jul 11: $0.70 Jul 12: $0.70 Jul 13: $0.70 Jul 14: $0.70 Jul 15: $0.70 Jul 16: $0.70 Jul 17: $0.70 Jul 18: $0.70 Jul 19: $0.70 Jul 20: $0.70 Jul 21: $0.70 Jul 22: $0.70 Jul 23: $0.70 Jul 24: $0.70 Jul 25: $0.70 Jul 26: $0.70 Jul 27: $0.70 Jul 28: $0.70 Jul 29: $0.70 Jul 30: $0.70 Jul 31: $0.70 Jun 15 Jul 31

Price war

Biggest recent price moves

When a provider changes a price, it shows up here. Green is a cut, red is a hike (blended $/Mtok).

mistral/mistral-medium-latest

Mistral · $0.8 → $3

+275%

sambanova/MiniMax-M2.7

Sambanova · $0.52 → $1.05

+100%

sambanova/gpt-oss-120b

Sambanova · $3.38 → $0.31

-90.7%

amazon.titan-embed-text-v2:0

Bedrock · $0.15 → $0.02

-90%

bedrock_mantle/openai.gpt-5.6-luna

Bedrock Mantle · $2.48 → $0.5

-80%

gpt-5.6-luna

Openai · $2.25 → $0.45

-80%

Intelligence & speed

Not just cheap, capable

Independent quality and speed benchmarks, so cost reads next to capability. Smartest, fastest, and the best capability per dollar.

🧠 Smartest

Intelligence Index

  1. 1 Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) Anthropic 60.1
  2. 2 GPT-5.5 (xhigh) OpenAI 54.8
  3. 3 Claude Sonnet 5 (Adaptive Reasoning, Max Effort) Anthropic 53.4
  4. 4 GPT-5.6 Luna (max) OpenAI 51.2
  5. 5 Claude Opus 5 (Adaptive Reasoning, Low Effort) Anthropic 50.6
  6. 6 GPT-5.5 (medium) OpenAI 50.4

⚡ Fastest

Output speed

  1. 1 Mercury 2 Inception 945 tok/s
  2. 2 Step 3.7 Flash StepFun 413 tok/s
  3. 3 LFM2.5-VL-1.6B Liquid AI 401 tok/s
  4. 4 HyperNova 60B 2605 Multiverse Computing 375 tok/s
  5. 5 Gemini 3.5 Flash-Lite Google 368 tok/s
  6. 6 Granite 3.3 8B (Non-reasoning) IBM 368 tok/s

💎 Best value

Intelligence per $/Mtok

  1. 1 HyperNova 60B 2605 Multiverse Computing 273.8 pts/$
  2. 2 Qwen3.5 4B (Non-reasoning) Alibaba 266.7 pts/$
  3. 3 Gemma 4 E4B (Non-reasoning) Google 222.5 pts/$
  4. 4 DeepSeek V4 Flash (Reasoning, High Effort) DeepSeek 213.7 pts/$
  5. 5 MiMo-V2.5 Xiaomi 212.6 pts/$
  6. 6 Step 3.5 Flash 2603 StepFun 173.3 pts/$

Quality & speed data from Artificial Analysis.

Human preference

What people actually prefer

The LMArena (Chatbot Arena) overall leaderboard, decided by millions of blind head-to-head votes. Snapshot Jul 21, 2026.

  • 1 claude-opus-4-6-thinking Anthropic 1503 63.2K votes
  • 2 gpt-5.4 Openai 1499 61.8K votes
  • 3 claude-opus-4-6 Anthropic 1497 67K votes
  • 4 gpt-5.4-mini-high Openai 1496 57.6K votes
  • 5 claude-fable-5 Anthropic 1493 14.6K votes
  • 6 gpt-5.2-high Openai 1493 47.4K votes
  • 7 gemini-3.6-flash Google 1491 4.7K votes
  • 8 gemini-3.5-flash-high Google 1490 10.1K votes
  • 9 claude-opus-4-7-thinking Anthropic 1490 50.7K votes
  • 10 gpt-5.5 Openai 1488 47K votes

Data: LMArena leaderboard, CC BY 4.0.

Real usage

What the community really spends

Anonymized, aggregate, opt-in. Numbers from real engineers, not list prices.

Tracked spend

$648.15

Tokens

67B

Tokens / $

103.3M

Contributors

5

Community dashboard →

Daily community spend

$0.00 $15.2 $30.5 $45.7 May 22: $6.98 May 23: $1.90 Jun 2: $12.9 Jun 3: $28.6 Jun 4: $11.5 Jun 10: $0.88 Jun 11: $5.14 Jun 12: $13.2 Jun 15: $18.1 Jun 16: $5.31 Jun 17: $0.74 Jun 18: $6.15 Jun 19: $28.9 Jun 20: $36.9 Jun 21: $0.15 Jun 22: $11.9 Jun 23: $7.57 Jun 25: $0.92 Jun 28: $0.43 Jun 30: $2.43 Jul 1: $1.35 Jul 2: $16.9 Jul 3: $18.2 Jul 5: $17.6 Jul 7: $20.9 Jul 8: $12.8 Jul 9: $4.87 Jul 11: $29.5 Jul 12: $20.3 Jul 13: $11.3 Jul 15: $8.49 Jul 16: $33.2 Jul 17: $40.8 Jul 18: $15.9 Jul 19: $24.3 Jul 21: $5.59 Jul 22: $17.8 Jul 23: $13.1 Jul 24: $21.6 Jul 26: $7.99 Jul 27: $11.1 Jul 28: $12.2 Jul 29: $9.60 Jul 30: $34.9 Jul 31: $37.3 May 22 Jul 31

By provider

  • Anthropic 100%

By platform

  • Claude Code 100%

By use case

  • Coding 100%

Value

Cheapest & priciest right now

Blended $/Mtok (3:1 input:output) across text models with public prices.

Cheapest

  • fireworks_ai/accounts/fireworks/models/flux-1-dev-controlnet-union Fireworks Ai $0/Mtok
  • fireworks-ai-embedding-up-to-150m Fireworks Ai Embedding Models $0.01/Mtok
  • fireworks-ai-embedding-150m-to-350m Fireworks Ai Embedding Models $0.01/Mtok
  • nscale/Qwen/Qwen2.5-Coder-3B-Instruct Nscale $0.02/Mtok
  • nscale/Qwen/Qwen2.5-Coder-7B-Instruct Nscale $0.02/Mtok
  • nebius/Qwen/Qwen2.5-Coder-7B Nebius $0.02/Mtok

Priciest

  • wandb/deepseek-ai/DeepSeek-R1-0528 Wandb $236,250/Mtok
  • wandb/deepseek-ai/DeepSeek-V3-0324 Wandb $154,250/Mtok
  • wandb/Qwen/Qwen3-Coder-480B-A35B-Instruct Wandb $112,500/Mtok
  • wandb/zai-org/GLM-4.5 Wandb $91,250/Mtok
  • wandb/deepseek-ai/DeepSeek-V3.1 Wandb $82,500/Mtok
  • wandb/meta-llama/Llama-3.3-70B-Instruct Wandb $71,000/Mtok

Explore

Every model, your way

Filter by provider, search, and sort the entire live price catalog. Pulled straight from the open Data API.

Model Provider Input Output Blended

Source: GET /api/v1/data/models · CC BY 4.0

The cheatsheet

Which AI for which job

No single model wins everything. The top pick for each job, ranked from real data.

🧠

Reasoning

Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

💻

Coding

Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

🤖

Agents

Claude Opus 5 (Adaptive Reasoning, Xhigh Effort)

💬

Chat & writing

claude-opus-4-6-thinking

⚡

Speed

Step 3.7 Flash

📚

Long context

Llama 4 Scout 17b 128e Instruct Maas

💎

Best value

Qwen3.5 4B (Non-reasoning)

🪙

Budget

Llama 3.1 8b

Full cheatsheet →

The price sheet lies

List price isn't what you pay

Qwen3.6 27B (Non-reasoning) is listed 88% cheaper than GPT-5.5 (Non-reasoning) — yet costs 85% more to actually run. Each model by list price (left) vs the real bill to run the Intelligence Index (right).

LIST PRICE $ per million tokens ACTUAL COST $ to run the Intelligence Index HyperNova 60B 2605$0.07 HyperNova 60B 2605 — list $0.07/Mtok HY Qwen3.5 9B (Reasoning)$0.16 Qwen3.5 9B (Reasoning) — list $0.16/Mtok QW gpt-oss-120b (high)$0.26 gpt-oss-120b (high) — list $0.26/Mtok GP GPT-5.6 Luna (max)$0.45 GPT-5.6 Luna (max) — list $0.45/Mtok GT MiMo-V2.5-Pro$0.54 MiMo-V2.5-Pro — list $0.54/Mtok MI Mistral Large 3$0.75 Mistral Large 3 — list $0.75/Mtok MS Qwen3.5 122B A10B (Non-…$1.10 Qwen3.5 122B A10B (Non-reasoning) — list $1.10/Mtok QE Qwen3.6 27B (Non-reason…$1.35 Qwen3.6 27B (Non-reasoning) — list $1.35/Mtok QN Claude 4.5 Haiku (Reaso…$2.00 Claude 4.5 Haiku (Reasoning) — list $2.00/Mtok CL Gemini 3.5 Flash (high)$3.38 Gemini 3.5 Flash (high) — list $3.38/Mtok GE GPT-5 (high)$3.44 GPT-5 (high) — list $3.44/Mtok G5 Kimi K3 (low)$6.00 Kimi K3 (low) — list $6.00/Mtok KI GPT-5.5 (Non-reasoning)$11.25 GPT-5.5 (Non-reasoning) — list $11.25/Mtok G2 GPT-5.5 (xhigh)$11.25 GPT-5.5 (xhigh) — list $11.25/Mtok G3 $37HyperNova 60B 2605 HyperNova 60B 2605 — actually $37 to run the Intelligence Index ($0.0193/task) HY $241Qwen3.5 9B (Reasoning) Qwen3.5 9B (Reasoning) — actually $241 to run the Intelligence Index ($0.2211/task) QW $96gpt-oss-120b (high) gpt-oss-120b (high) — actually $96 to run the Intelligence Index ($0.0607/task) GP $191GPT-5.6 Luna (max) GPT-5.6 Luna (max) — actually $191 to run the Intelligence Index ($0.0571/task) GT $108MiMo-V2.5-Pro MiMo-V2.5-Pro — actually $108 to run the Intelligence Index ($0.0433/task) MI $71Mistral Large 3 Mistral Large 3 — actually $71 to run the Intelligence Index ($0.0599/task) MS $219Qwen3.5 122B A10B (No… Qwen3.5 122B A10B (Non-reasoning) — actually $219 to run the Intelligence Index ($0.1793/task) QE $357Qwen3.6 27B (Non-reas… Qwen3.6 27B (Non-reasoning) — actually $357 to run the Intelligence Index ($0.3602/task) QN $539Claude 4.5 Haiku (Rea… Claude 4.5 Haiku (Reasoning) — actually $539 to run the Intelligence Index ($0.2375/task) CL $1,041Gemini 3.5 Flash (hig… Gemini 3.5 Flash (high) — actually $1,041 to run the Intelligence Index ($0.5855/task) GE $793GPT-5 (high) GPT-5 (high) — actually $793 to run the Intelligence Index ($0.2215/task) G5 $283Kimi K3 (low) Kimi K3 (low) — actually $283 to run the Intelligence Index ($0.1885/task) KI $193GPT-5.5 (Non-reasonin… GPT-5.5 (Non-reasoning) — actually $193 to run the Intelligence Index ($0.1515/task) G2 $2,778GPT-5.5 (xhigh) GPT-5.5 (xhigh) — actually $2,778 to run the Intelligence Index ($0.9925/task) G3
Pricier than it looks Secret bargain
Full breakdown →

Run cost = cost to run the full Intelligence Index, from Artificial Analysis.

Citation

Use this in your work

Quote any figure on this page. It is all open data. Here is a ready-made citation, and the methodology behind every number.

Copy a citation

Free to use and cite under CC BY 4.0. See how this is measured.

APA

Champlin Enterprises. (2026). The State of AI (MyTokenTracker) [Data set]. MyTokenTracker. Retrieved July 31, 2026, from https://mytokentracker.io/state-of-ai

BibTeX 
@misc{mytokentracker-state-of-ai,
  title        = {The State of AI (MyTokenTracker)},
  author       = {{Champlin Enterprises}},
  year         = {2026},
  howpublished = {MyTokenTracker, \url{https://mytokentracker.io/state-of-ai}},
  note         = {Accessed July 31, 2026. Licensed CC BY 4.0.},
  url          = {https://mytokentracker.io/state-of-ai}
}

Need a fixed point in time? Every day’s data is permanently archived in the open-data repository, so you can cite a specific date by linking that day’s committed file.

Free weekly digest

Get this in your inbox, weekly

The State of AI moves every day. Once a week we send the headline: the AI Cost Index, the biggest price moves, and what it means. Free, no account.

No spam, no account. One click to leave.

Measure it. Don't guess.

Every number on this page came from people who track instead of guess. One line to install, automatic capture, free forever. Add your usage and the whole picture gets sharper.

Start free Take the data

Our wiggly friend is an original MyTokenTracker mascot, here purely for fun. It is not affiliated with, endorsed by, or representing Anthropic, OpenAI, Google, or any model provider. Product and model names (Claude, GPT, Gemini, and others) are trademarks of their respective owners and appear here only to report public pricing and usage.