index · updated 2026-08-23 · since 2026-07-17

The Token Price Index

What a million AI tokens costs, tracked every day. The Market line is the median blended price across every paid model in the catalog; the Frontier line is the same median over today's twenty highest-scoring models. Prices are blended 3:1 input to output, the usual shape of real workloads.

Market · median of 411 paid models
$0.88/ 1M tokens
+5.4% 30 days +7.7% since 2026-07-17
Frontier 20 · median of the top-scoring models
$3.22/ 1M tokens
-42.8% 30 days -42.8% since 2026-07-17
Last 30 days
34 cuts 17 increases
37 trading days in the series
MarketFrontier 20USD per 1M tokens, blended
$0.23$2.22$4.21$6.20 Jul 17Aug 5Aug 23

Yesterday's moves 1 new, 10 price changes

ModelWasNowChange
DeepSeek: DeepSeek V4 Flash 0423 $0.094$0.066 -29.7%
DeepSeek V4 Flash Latest $0.094$0.076 -18.7%
DeepSeek: DeepSeek V4 Pro 0423 $0.623$0.517 -17.0%
Qwen: Qwen3.8 27B $1.138$1.050 -7.7%
DeepSeek: DeepSeek V4 Pro 0813 $1.782$1.683 -5.6%
Qwen: Qwen3 30B A3B $0.228$0.215 -5.5%
Thinking Machines: Inkling $1.725$1.762 +2.2%
OpenAI: gpt-oss-120b $0.065$0.070 +8.1%
Meta: Muse Glimmer 30B $0.500$0.637 +27.5%
MoonshotAI: Kimi K2.6 $0.976$1.712 +75.4%

Biggest cuts, 30 days

ModelWasNowChange
OpenAI: GPT-5.6 Lunasince 2026-08-18 $2.250$0.450-80.0%
OpenAI: GPT-5.6 Luna Prosince 2026-08-18 $2.250$0.450-80.0%
Google Gemini Flash Latestsince 2026-08-14 $3.000$0.750-75.0%
OpenAI: GPT-5.6 Solsince 2026-08-22 $11.250$4.000-64.4%
OpenAI: GPT-5.6 Sol Prosince 2026-08-22 $11.250$4.000-64.4%
OpenAI GPT Latestsince 2026-08-22 $11.250$4.000-64.4%
Qwen: Qwen3 14Bsince 2026-08-10 $0.398$0.150-62.3%
Google: Gemini 3.6 Flashsince 2026-08-14 $3.000$1.500-50.0%
DeepSeek: DeepSeek V4 Flash 0423since 2026-08-23 $0.123$0.066-45.8%
Qwen: Qwen3 30B A3B Instruct 2507since 2026-07-26 $0.150$0.084-43.8%
Z.ai: GLM 5since 2026-08-15 $1.350$0.930-31.1%
Qwen: Qwen3 Coder 480B A35Bsince 2026-07-25 $0.673$0.475-29.4%

Biggest increases, 30 days

ModelWasNowChange
Tencent: Hy3 previewsince 2026-08-16 $0.100$0.285+185.7%
Qwen: Qwen2.5 7B Instructsince 2026-07-30 $0.055$0.125+127.3%
DeepSeek: DeepSeek V3.1since 2026-08-22 $0.425$0.825+94.1%
Qwen: Qwen3.6 27Bsince 2026-08-20 $0.816$1.350+65.4%
Qwen: Qwen3 30B A3B Thinking 2507since 2026-07-30 $0.488$0.750+53.8%
Qwen: Qwen3 Next 80B A3B Thinkingsince 2026-07-30 $0.268$0.412+53.8%
Qwen: Qwen3 VL 235B A22B Thinkingsince 2026-08-07 $0.845$1.300+53.8%
Qwen: Qwen3 VL 30B A3B Thinkingsince 2026-07-30 $0.488$0.750+53.8%

How the index is computed

Blended price. For every model, the blended cost per million tokens is three parts input price to one part output price, the typical ratio of a production workload. Free models and zero-priced listings are excluded.

Market. The median blended price across every paid model in the catalog that day. A median, not a mean, so a single $75-per-million flagship or a $0.02 distillate cannot drag it around. New models enter the day they appear; retired models stay at their last price until the feed drops them.

Frontier 20. The same median over today's twenty highest quality-score models in the catalog, fixed as a basket and priced back through history: Claude Opus 5, SpaceXAI: Grok 4.6, OpenAI: GPT-5.6 Sol, Z.ai: GLM 5.3, Anthropic: Claude Fable 5, Qwen: Qwen3.8 Max, Qwen: Qwen3.8 2.4T A95B, MoonshotAI: Kimi K3, Anthropic: Claude Opus 4.8, Google: Gemini 2.5 Pro, OpenAI: GPT-5.5, OpenAI: GPT-5.6 Terra, Meta: Muse Spark 1.2, Anthropic: Claude Sonnet 5, SpaceXAI: Grok 4.5, DeepSeek: DeepSeek V4 Pro 0423, OpenAI: GPT-5.4, Qwen: Qwen3.8 27B, Google: Gemini 3.7 Flash, Z.ai: GLM 5.2.

History. Prices are recorded from the OpenRouter catalog on every daily rebuild since 2026-07-17 and kept as a change log, so each day's value is the last price seen on or before that date. The movers tables compare each model's price today with its price thirty days earlier. The whole series is available as tpi.json; for live per-model pricing, see the API.

Rebuilt 2026-08-23. Context, not financial advice; vendor list prices, before volume discounts.