Skip to main content

Gradually LLM Price Index

What do LLM tokens really cost?

The Gradually LLM Price Index (GLLMPI) makes API prices comparable. You get a transparent market median from a curated selection of publicly priced text and code models with dynamic price bands based on the currently displayed prices.

214 models and variants30 provider groups18 source systems18,740 checked price points15 pricing catalogs

Current market median

Methodology v2.8

$0.658

per 1M tokens

Up to $1 · 132
$0.263
Over $1 to $5 · 59
$1.57
Over $5 · 23
$8.00

Source pages checked on September 19, 2026. Prices checked on September 19, 2026. 75% input, 25% output.

Historical index

LLM token price history

The series combines verified historical price checkpoints since September 2022 with daily market observations since August 2026. The first historical checkpoint equals 100 index points.

Input and output prices are weighted by your token mix.

Share of input tokens

75%
75% Input25% Output

Open models includes open-source and open-weights models.

Current level

52.60

-47.40 · -47.40%cheaper since inception
Market price per 1M tokens$0.658
High787.40
Low34.00
Historical LLM Price Index chart
DateLLM Price IndexMarket price per 1M tokensData basis
September 1, 2022100.00$1.25Historically reconstructed
March 1, 2023160.00$2.00Historically reconstructed
March 14, 2023160.00$2.00Historically reconstructed
May 10, 2023460.70$5.76Historically reconstructed
June 13, 2023510.70$6.38Historically reconstructed
July 11, 2023787.40$9.84Historically reconstructed
August 11, 2023146.00$1.83Historically reconstructed
September 28, 2023144.00$1.80Historically reconstructed
November 6, 2023144.00$1.80Historically reconstructed
November 13, 2023144.00$1.80Historically reconstructed
November 21, 2023144.00$1.80Historically reconstructed
November 28, 2023144.00$1.80Historically reconstructed
November 29, 2023112.00$1.40Historically reconstructed
November 30, 2023112.00$1.40Historically reconstructed
December 1, 2023112.00$1.40Historically reconstructed
January 4, 2024112.00$1.40Historically reconstructed
January 25, 2024112.00$1.40Historically reconstructed
February 15, 2024144.00$1.80Historically reconstructed
February 21, 202480.00$1.00Historically reconstructed
February 26, 202480.00$1.00Historically reconstructed
March 4, 202480.00$1.00Historically reconstructed
March 5, 202480.00$1.00Historically reconstructed
March 11, 202460.00$0.75Historically reconstructed
March 13, 202460.00$0.75Historically reconstructed
April 4, 202484.00$1.05Historically reconstructed
April 11, 202484.00$1.05Historically reconstructed
May 6, 202468.50$0.856Historically reconstructed
May 7, 202472.00$0.90Historically reconstructed
May 13, 202472.00$0.90Historically reconstructed
May 29, 202477.50$0.969Historically reconstructed
May 30, 202495.00$1.19Historically reconstructed
June 1, 202477.50$0.969Historically reconstructed
June 21, 202477.50$0.969Historically reconstructed
July 18, 202472.00$0.90Historically reconstructed
July 24, 202477.50$0.969Historically reconstructed
August 6, 202477.50$0.969Historically reconstructed
August 12, 202477.50$0.969Historically reconstructed
August 22, 202495.00$1.19Historically reconstructed
August 30, 202480.00$1.00Historically reconstructed
September 5, 202480.00$1.00Historically reconstructed
September 12, 202480.00$1.00Historically reconstructed
September 13, 202480.00$1.00Historically reconstructed
September 17, 202460.00$0.75Historically reconstructed
September 27, 202460.00$0.75Historically reconstructed
October 1, 202460.00$0.75Historically reconstructed
October 14, 202448.00$0.60Historically reconstructed
October 16, 202448.00$0.60Historically reconstructed
October 22, 202448.00$0.60Historically reconstructed
November 6, 202448.00$0.60Historically reconstructed
December 3, 202448.00$0.60Historically reconstructed
December 10, 202448.00$0.60Historically reconstructed
December 12, 202454.00$0.675Historically reconstructed
December 13, 202439.00$0.488Historically reconstructed
December 17, 202439.00$0.488Historically reconstructed
December 26, 202439.00$0.488Historically reconstructed
January 15, 202534.00$0.425Historically reconstructed
January 20, 202545.40$0.568Historically reconstructed
January 31, 202545.40$0.568Historically reconstructed
February 3, 202545.40$0.568Historically reconstructed
February 5, 202545.40$0.568Historically reconstructed
February 8, 202548.00$0.60Historically reconstructed
February 24, 202548.00$0.60Historically reconstructed
February 28, 202548.00$0.60Historically reconstructed
March 13, 202557.50$0.719Historically reconstructed
March 16, 202558.75$0.734Historically reconstructed
March 19, 202558.75$0.734Historically reconstructed
April 4, 202558.75$0.734Historically reconstructed
April 14, 202558.75$0.734Historically reconstructed
April 16, 202558.75$0.734Historically reconstructed
April 17, 202558.75$0.734Historically reconstructed
April 24, 202560.00$0.75Historically reconstructed
April 28, 202568.50$0.856Historically reconstructed
May 7, 202568.50$0.856Historically reconstructed
May 21, 202567.25$0.841Historically reconstructed
May 22, 202567.25$0.841Historically reconstructed
June 6, 202567.25$0.841Historically reconstructed
June 10, 202567.25$0.841Historically reconstructed
June 15, 202557.50$0.719Historically reconstructed
June 16, 202557.50$0.719Historically reconstructed
June 17, 202557.50$0.719Historically reconstructed
July 10, 202557.50$0.719Historically reconstructed
July 14, 202557.50$0.719Historically reconstructed
July 15, 202557.50$0.719Historically reconstructed
July 18, 202554.25$0.678Historically reconstructed
July 21, 202554.25$0.678Historically reconstructed
July 22, 202554.25$0.678Historically reconstructed
July 28, 202554.25$0.678Historically reconstructed
July 30, 202554.25$0.678Historically reconstructed
August 1, 202557.50$0.719Historically reconstructed
August 5, 202557.50$0.719Historically reconstructed
August 7, 202557.50$0.719Historically reconstructed
August 11, 202554.25$0.678Historically reconstructed
August 12, 202554.25$0.678Historically reconstructed
August 28, 202554.25$0.678Historically reconstructed
September 1, 202554.25$0.678Historically reconstructed
September 15, 202554.25$0.678Historically reconstructed
September 19, 202549.50$0.619Historically reconstructed
September 29, 202549.50$0.619Historically reconstructed
October 6, 202549.50$0.619Historically reconstructed
October 15, 202551.00$0.638Historically reconstructed
October 27, 202545.70$0.571Historically reconstructed
October 28, 202545.70$0.571Historically reconstructed
November 6, 202545.70$0.571Historically reconstructed
November 13, 202545.70$0.571Historically reconstructed
November 19, 202545.70$0.571Historically reconstructed
November 24, 202545.70$0.571Historically reconstructed
December 2, 202545.70$0.571Historically reconstructed
December 11, 202545.70$0.571Historically reconstructed
January 5, 202645.70$0.571Historically reconstructed
January 26, 202644.35$0.554Historically reconstructed
February 5, 202644.35$0.554Historically reconstructed
February 12, 202644.35$0.554Historically reconstructed
February 17, 202644.35$0.554Historically reconstructed
February 19, 202644.35$0.554Historically reconstructed
March 5, 202644.35$0.554Historically reconstructed
March 16, 202644.35$0.554Historically reconstructed
March 17, 202644.35$0.554Historically reconstructed
March 26, 202644.35$0.554Historically reconstructed
April 16, 202644.35$0.554Historically reconstructed
April 20, 202644.35$0.554Historically reconstructed
April 24, 202644.35$0.554Historically reconstructed
April 28, 202644.35$0.554Historically reconstructed
May 15, 202651.60$0.645Historically reconstructed
May 28, 202651.60$0.645Historically reconstructed
May 29, 202651.60$0.645Historically reconstructed
June 1, 202662.75$0.784Historically reconstructed
June 2, 202664.00$0.80Historically reconstructed
June 15, 202664.00$0.80Historically reconstructed
June 18, 202668.00$0.85Historically reconstructed
June 25, 202668.00$0.85Historically reconstructed
June 30, 202668.00$0.85Historically reconstructed
July 9, 202668.00$0.85Historically reconstructed
July 15, 202668.00$0.85Historically reconstructed
July 16, 202668.00$0.85Historically reconstructed
July 30, 202668.00$0.85Historically reconstructed
August 1, 202668.00$0.85Historically reconstructed
August 3, 202668.00$0.85Historically reconstructed
August 5, 202668.00$0.85Historically reconstructed
August 11, 202664.00$0.80Historically reconstructed
August 12, 202664.00$0.80Historically reconstructed
August 13, 202677.00$0.963Historically reconstructed
August 14, 2026105.25$1.32Daily observation
August 15, 2026105.25$1.32Daily observation
August 16, 2026110.75$1.38Daily observation
August 17, 2026110.75$1.38Daily observation
August 18, 2026110.75$1.38Daily observation
August 19, 2026110.75$1.38Daily observation
August 20, 2026110.75$1.38Daily observation
August 21, 202677.00$0.963Historically reconstructed
August 22, 202642.00$0.525Daily observation
August 23, 202642.00$0.525Daily observation
August 24, 202650.20$0.628Daily observation
August 25, 202650.20$0.628Daily observation
August 26, 202650.20$0.628Daily observation
August 27, 202648.02$0.60Daily observation
August 28, 202650.20$0.628Daily observation
August 29, 202648.02$0.60Daily observation
August 30, 202677.00$0.963Historically reconstructed
August 31, 202677.00$0.963Historically reconstructed
September 1, 202677.00$0.963Historically reconstructed
September 2, 202648.02$0.60Daily observation
September 3, 202648.02$0.60Daily observation
September 4, 202648.02$0.60Daily observation
September 5, 202648.02$0.60Daily observation
September 6, 202648.02$0.60Daily observation
September 7, 202648.02$0.60Daily observation
September 8, 202648.02$0.60Daily observation
September 9, 202690.00$1.13Historically reconstructed
September 10, 202690.00$1.13Historically reconstructed
September 11, 202690.00$1.13Historically reconstructed
September 12, 202649.01$0.613Daily observation
September 13, 202649.01$0.613Daily observation
September 14, 202649.01$0.613Daily observation
September 15, 202649.01$0.613Daily observation
September 16, 202649.01$0.613Daily observation
September 17, 202649.01$0.613Daily observation
September 18, 202652.60$0.658Daily observation
September 19, 202652.60$0.658Daily observation

The historical index begins on September 1, 2022. Each checkpoint uses the model generation available at that time. Values since August 14, 2026 come from the automated daily refresh.

Task costs

What does one coding benchmark task cost?

These published coding measurements spread each configuration’s total API cost estimate across all 89 tasks in the benchmark.

Published measurementsCoding · 89 tasks · 445 published trials in total

Page-wide filters select current model families. Input, output, and token mix still determine current token prices and price bands, but do not reprice these published test costs.

9 model families with 12 published configurations in your selection. Comparable measurements are not available here for the other selected models.

Lower published costs appear first.

Grok 4.5

Cursor CLI · high

$1.51

Per benchmark task

Tasks 89Runs 445

Gemini 3.1 Pro

Terminus 2 · high

$2.58

Per benchmark task

Tasks 89Runs 445

Gemini 3.1 Pro

Gemini CLI · high

$2.66

Per benchmark task

Tasks 89Runs 445

GPT-5.6 Luna

Codex · max

$2.71

Per benchmark task

Tasks 89Runs 445

GLM-5.1

Claude Code · max

$3.11

Per benchmark task

Tasks 89Runs 445

Claude Sonnet 5

Claude Code · high

$3.24

Per benchmark task

Tasks 89Runs 445

GPT-5.6 Terra

Codex · max

$4.73

Per benchmark task

Tasks 89Runs 445

Claude Fable 5

Terminus 2 · high

$4.93

Per benchmark task

Tasks 89Runs 445

GPT-5.5

Terminus 2 · xhigh

$5.55

Per benchmark task

Tasks 89Runs 445

Claude Fable 5

Claude Code · xhigh

$6.21

Per benchmark task

Tasks 89Runs 445

Claude Opus 4.7

Terminus 2 · max

$6.54

Per benchmark task

Tasks 89Runs 445

GPT-5.5

Codex · xhigh

$23.14

Per benchmark task

Tasks 89Runs 445
Inspect the test protocol and calculation

We divide the total published API cost estimate for a configuration by the benchmark’s 89 tasks. The dataset reports 445 trials in total, but does not disclose their task-level distribution or cost telemetry. When cost telemetry is missing for a trial, the published total may count it as $0. We therefore do not present a supposedly exact cost per successfully solved task.

Published API cost per coding benchmark task in USD
Model and configurationTasksRunsTotal API costPer benchmark task
Grok 4.5Cursor CLI · high89445$134.09$1.51
Gemini 3.1 ProTerminus 2 · high89445$229.99$2.58
Gemini 3.1 ProGemini CLI · high89445$236.49$2.66
GPT-5.6 LunaCodex · max89445$241.45$2.71
GLM-5.1Claude Code · max89445$277.14$3.11
Claude Sonnet 5Claude Code · high89445$288.18$3.24
GPT-5.6 TerraCodex · max89445$421.15$4.73
Claude Fable 5Terminus 2 · high89445$438.64$4.93
GPT-5.5Terminus 2 · xhigh89445$493.85$5.55
Claude Fable 5Claude Code · xhigh89445$552.67$6.21
Claude Opus 4.7Terminus 2 · max89445$582.26$6.54
GPT-5.5Codex · xhigh89445$2,059.19$23.14

Published costs are estimates for the respective configuration, not audited invoices. The model, agent, reasoning budget, and environment differ. Results apply to these coding tasks, not arbitrary workflows. Human review and additional operating costs are not included.

Benchmark data as of September 3, 2026.

Data retrieved September 19, 2026.

Forecast

How LLM API prices could evolve

Three paths with declining prices and a separate cost stress case begin with the latest verified median of your selection and extend through September 19, 2029. They show possible market paths, not a pricing commitment.

Latest verified median of your selection

$0.658

per 1M tokens

Horizon
36 months
Price index, base value = 100
100
Anchor date
September 19, 2026
The global filters determine the latest verified anchor rate and every displayed dollar amount. The scenarios intentionally remain consistent market paths.

Scenario range

Slower price decline

20% per year

$0.34

September 19, 2029

Base scenario

40% per year

$0.14

September 19, 2029

Faster price decline

60% per year

$0.04

September 19, 2029

A forecast is not a price commitment. New model generations, hardware costs, competition, demand, and pricing policy can materially change the outcome.

Stress scenario

Cost and capacity stress

This stress case is not a prediction. It assumes persistently high demand for advanced compute, constrained power and accelerator capacity, and greater pass-through of infrastructure and financing costs.

Assumption
API prices rising by 35% per year
Price after 36 months
2.46× · $1.62

A correction in AI capital markets would not automatically raise prices. Weaker demand and excess capacity could push them lower. The upward path requires demand to remain high while capital becomes more expensive and capacity stays constrained.

Providers do not publish per-model unit costs or pricing margins. The scenario therefore does not claim that today's API prices are subsidized.

Why this range?

API prices for a fixed capability level have sometimes fallen by orders of magnitude. The GLLMPI, however, covers the entire market, including new premium models. The forecast therefore uses much more cautious annual declines of 20%, 40%, and 60%.

What pushes prices down
More compute per watt, smaller models with comparable quality, sparse architectures, quantization, better utilization, and faster decoding reduce cost per token. Competition increases the pressure to pass those gains on.
What can slow the decline
Longer contexts, reasoning, and agent runs consume more input and output tokens. Scarce data center capacity, rising electricity demand, high demand, and premium pricing can delay price reductions.
What the chart forecasts
The base API price per 1M tokens for the current selection. It does not forecast total cost per task, cache, batch, volume, or contract pricing, and it is not a guarantee for individual models.
Scenario forecast for LLM API prices
DateSlower price declineBase scenarioFaster price declineCost and capacity stress
September 19, 2026$0.66$0.66$0.66$0.66
October 19, 2026$0.65$0.63$0.61$0.67
November 19, 2026$0.63$0.60$0.56$0.69
December 19, 2026$0.62$0.58$0.52$0.71
January 19, 2027$0.61$0.55$0.48$0.73
February 19, 2027$0.60$0.53$0.45$0.75
March 19, 2027$0.59$0.51$0.42$0.76
April 19, 2027$0.58$0.49$0.39$0.78
May 19, 2027$0.57$0.47$0.36$0.80
June 19, 2027$0.56$0.45$0.33$0.82
July 19, 2027$0.55$0.43$0.31$0.84
August 19, 2027$0.54$0.41$0.28$0.87
September 19, 2027$0.53$0.39$0.26$0.89
October 19, 2027$0.52$0.38$0.24$0.91
November 19, 2027$0.51$0.36$0.23$0.93
December 19, 2027$0.50$0.35$0.21$0.96
January 19, 2028$0.49$0.33$0.19$0.98
February 19, 2028$0.48$0.32$0.18$1.01
March 19, 2028$0.47$0.31$0.17$1.03
April 19, 2028$0.46$0.29$0.15$1.06
May 19, 2028$0.45$0.28$0.14$1.08
June 19, 2028$0.44$0.27$0.13$1.11
July 19, 2028$0.44$0.26$0.12$1.14
August 19, 2028$0.43$0.25$0.11$1.17
September 19, 2028$0.42$0.24$0.11$1.20
October 19, 2028$0.41$0.23$0.10$1.23
November 19, 2028$0.41$0.22$0.09$1.26
December 19, 2028$0.40$0.21$0.08$1.29
January 19, 2029$0.39$0.20$0.08$1.32
February 19, 2029$0.38$0.19$0.07$1.36
March 19, 2029$0.38$0.18$0.07$1.39
April 19, 2029$0.37$0.18$0.06$1.43
May 19, 2029$0.36$0.17$0.06$1.46
June 19, 2029$0.36$0.16$0.05$1.50
July 19, 2029$0.35$0.15$0.05$1.54
August 19, 2029$0.34$0.15$0.05$1.58
September 19, 2029$0.34$0.14$0.04$1.62

Price range

How far apart the models in the index really are

Each bar shows the blended list price per 1M tokens at the selected input share.

214 models match the current filters.

GPT-5.4 Pro

OpenAIUnited StatesOver $5

$67.50

GPT-5.5 Pro

OpenAIUnited StatesOver $5

$67.50

GPT-5 pro

OpenAIUnited StatesOver $5

$41.25

o3-pro

OpenAIUnited StatesOver $5

$35.00

o1

OpenAIUnited StatesOver $5

$26.25

Claude Fable 5

AnthropicUnited StatesOver $5

$20.00

Claude Fable 5.1

AnthropicUnited StatesOver $5

$20.00

Claude Mythos 5

AnthropicUnited StatesOver $5

$20.00

Claude Mythos 5.1

AnthropicUnited StatesOver $5

$20.00

GPT-6 Astra

OpenAIUnited StatesOver $5

$20.00

ChatGPT chat-latest

OpenAIUnited StatesOver $5

$11.25

GPT-5.5

OpenAIUnited StatesOver $5

$11.25

Claude Opus 4.5

AnthropicUnited StatesOver $5

$10.00

Claude Opus 4.6

AnthropicUnited StatesOver $5

$10.00

Claude Opus 4.7

AnthropicUnited StatesOver $5

$10.00

Claude Opus 5

AnthropicUnited StatesOver $5

$10.00

GPT-5.6 Sol

OpenAIUnited StatesOver $5

$8.80

Claude Sonnet 4.5

AnthropicUnited StatesOver $5

$6.00

Up to $1Over $1 to $5Over $5Median $0.658

Country comparison

Price by country

Each country value is the median of its provider medians, so a provider with many model variants does not count more heavily.

Country attribution follows the model developer’s headquarters, not the location of the data center.

Canada

4 models, 1 provider

Median price

$2.32

United States

97 models, 14 providers

Median price

$0.85

China

96 models, 13 providers

Median price

$0.628

South Korea

4 models, 1 provider

Median price

$0.263

France

13 models, 1 provider

Median price

$0.15

What the number means

A market benchmark you can recalculate

The index is deliberately not a mysterious financial ticker. You can inspect every model, price, and calculation rule. Change the token mix and the benchmark moves for a reason you can see.

Lowest-cost model
$0.025

Mistral Nemo

Highest-cost model
$67.50

GPT-5.4 Pro

Price multiple
2,700×

between the lowest-cost and highest-cost model

View every model and price

Verification scope

Checked daily, published in aggregate

The exact source composition remains editorial know-how. Scope, freshness, methodology, and validation rules remain publicly verifiable.

source systems
18
checked price points
18,740
Last checked
on September 19, 2026

The index covers 214 selected models from 15 pricing catalogs and 30 provider groups. Only complete input and output observations count.

Methodology

How the Gradually LLM Price Index is calculated

The calculation is intentionally simple. You can verify the benchmark yourself and see immediately what it captures and what it leaves out.

  1. 01

    Define the model selection

    The current index includes publicly priced text and code models available through an active developer API or API provider and marked as current in Gradually's central pricing database. Each historical checkpoint uses a time-specific model selection available at that date. If a verified direct developer price is unavailable for a historical model, a dated public cloud or hosting tariff with a complete input and output price pair may be used as an exception. Specialized image, audio, embedding, and moderation models are excluded.

  2. 02

    Load current token prices

    The daily job collects several concrete input and output tariffs for each model. It selects one genuinely available price pair near the median of trusted primary, gateway, and independent sources. For time-based tariffs, the regular input and output prices are the off-peak baseline. The index does not switch by the current clock time, so current and historical points stay comparable. Community catalogs act as cross-checks. Nothing is published if a selected model has no price at all.

  3. 03

    Show price bands

    Price bands use the currently displayed price per 1M tokens. Up to $1 is low, over $1 to $5 is mid, and over $5 is high. Switching between input, output, and blended prices recalculates every model's band.

  4. 04

    Balance providers fairly

    A separate median is calculated for each provider first. The overall index is then calculated from those provider medians, so a provider with many variants does not carry more weight.

What the index does not capture

The index uses the public base tier of API list prices. For time-based tariffs, that base tier is the off-peak rate. The index does not switch in real time by the clock, so current and historical points remain comparable on the same basis. It excludes context-based pricing tiers, regional price differences, negotiated discounts, caching, batch processing, request fees, search queries, citation and reasoning-token surcharges, self-hosted models, taxes, and real provider usage shares. Historical provider coverage is narrower before May 2024. Changes in the available model composition can also affect the series. The curated selection covers major providers, but not every API and hosting service in the market. The observed index series is a comparison benchmark, not a financial index. The separate forecast section shows scenarios only.

Want to calculate your actual API costs?

The index shows the market. For one specific model and your exact workflow, the full API cost calculator is the better tool.

Open the API cost calculatorCount tokens first

Frequently asked questions about the LLM Price Index

Short, honest answers

Changelog

The latest updates and improvements to our Gradually LLM Price Index
v2.26September 3, 2026

Muse Spark 1.3 and Qwen models refreshed

  • Added Muse Spark 1.3 to the current market selection in place of the superseded 1.2 version
  • Updated Qwen 3.8 Max to the 0902 snapshot while retaining the current global Alibaba rates
  • Added Qwen 3.8 Flash to the small segment with Alibaba's official tariff and separately observed market rates
  • Avoided treating Qwen 3.8 Flash Next as a second API tariff because QwenCloud serves the same open weights as Qwen3.8 Flash
  • Kept Gemini 3.8 Flash Cyber out of price statistics because no public API price is available
  • Refreshed official, router, and market sources for 217 current models
v2.24September 2, 2026

Daily refresh safeguards strengthened

  • Excluded models that disappeared from public API availability from the current market selection
  • Added structured fallback tariffs for KAT Coder models and restored the complete source refresh
  • Closed three missed daily snapshots with the last fully verified market state and labelled them as reconstructed
  • Added task costs to the public deployment fingerprint and independent freshness check
  • Disclosed the limitation of missing cost telemetry in published task-cost measurements
v2.23September 1, 2026

Current benchmark-cost coverage expanded

  • Replaced the previous historical selection with 12 current agent and model configurations across 9 model families
  • Distributed each published total cost transparently across the benchmark’s 89 coding tasks
  • Documented the 445 published trials per configuration without claiming an unsupported task-level distribution
  • Disclosed missing cost telemetry as a limitation of the published totals
  • Added a daily, failure-tolerant refresh with a dedicated failure alert for the supplementary task-cost snapshot
v2.22August 31, 2026

Cost per successfully solved task

  • Added historical coding runs with API cost per solved task and the corresponding success rate
  • Used actual success counts rather than rounded percentages in the calculation
  • Separated tested configurations, test dates, and cost estimates from current token prices
  • Connected model selection to the page-wide filters and explicitly identified missing test coverage
v2.21August 28, 2026

Added a scenario forecast

  • Added a new forecast section with three-year scenario charts
  • Shows slower, base, and faster price declines at 20%, 40%, and 60% per year as separate scenarios
  • Added cost and capacity stress at 35% annual price growth as a separate, explicitly non-consensus stress case
  • Global filters now determine the latest verified anchor rate and every projected dollar amount
  • Added a scenario range, a screen-reader table, and a clear uncertainty notice
  • The methodology now covers historical price trends, hardware efficiency, serving methods, and countervailing cost drivers
v2.20August 24, 2026

Market and historical coverage expanded

  • Added 20 current models and variants across additional families
  • The current selection now covers 213 models and variants, including 131 open models across 31 provider groups
  • Expanded the historical reconstruction to 142 verified price checkpoints since September 2022
  • Added historical Seed, Qwen, Solar, Grok, and Gemini tariffs backed by dated primary evidence
  • Separated the global Qwen 3.8 Max tariff cleanly from regional tariffs
  • Corrected the direct Inkling Small tariff and the current Muse Glimmer 30B prices
  • Daily validation now covers 213 models, 878 router tariffs, and more than 15,700 complete market prices
  • Added local provider marks for Meituan and Thinking Machines Lab
v2.19August 23, 2026

Historical index, filters, and FAQ corrected

  • The Open models selection now shows the matching historical data instead of an empty state
  • The Max view reflects verified market and price movements instead of remaining at 100 index points
  • Provider, country, access, and price-band filters now also affect the historical series
  • Added historical model types as well as legacy providers and their country attribution
  • The custom range now begins and ends at the first and last available checkpoint for the selection
  • The current selection covers 193 models and variants, including 125 open models across 27 provider groups
  • Added regression coverage for German, English, desktop, and mobile views
  • The FAQ now uses the same accessible accordion component as the LLM comparison
  • Removed the redundant formula block from the methodology section
v2.18August 22, 2026

Filters, country comparison, and open-weight coverage expanded

  • Filter providers individually or in combination using provider icons
  • Select open and proprietary models separately
  • Filter and sort models by the country where each provider is headquartered
  • Added a dedicated section with a provider-balanced median price for each country
  • Country values respond to price view, price band, provider, and model access filters
  • Added 106 open-weight models across major model families
  • The current selection now covers 190 models and variants, including 125 open models
  • Added active, older, multimodal, reasoning, and coding variants that still have available API tariffs
  • Four additional structured source systems increase the number of price points checked every day
  • Added Meta, Microsoft, and NVIDIA with country attribution and local provider icons
  • Price bands, sorting, and the model list stay consistent for every selection
v2.10August 17, 2026

Daily source and historical coverage expanded

  • 15 source systems now provide usable daily price observations
  • Removed an unnecessary configuration requirement for publicly available pricing data
  • Added 114 dated reconstructed index checkpoints since March 2023
  • Captured 149 verified price intervals across 18 provider groups
  • Refined early pricing phases and documented model retirements
v2.8August 16, 2026

Historical price series expanded and refined

  • Added 97 dated reconstructed index checkpoints since March 2023
  • Added more model generations and provider groups with verified USD price pairs
  • Captured price starts, price changes, and model retirements as separate events
  • Removed unsupported backdating and corrected event dates
  • Modelled price changes as separate, non-overlapping validity intervals
  • Clearly separated reconstructed values from daily observed values
  • Added long-term views for the overall index and the model classes used at the time
v2.4August 14, 2026

Daily market observations introduced

  • Added separate histories for input, output, and blended token prices
  • Added multi-stage price verification and automated publication
v2.0August 2, 2026

Daily refresh automated

  • Scheduled price verification twice per day
  • Added validation and production checks after each refresh
v1.0August 1, 2026

LLM Price Index launched

  • Launched the overall index and the three model classes used at the time
  • Published input, output, and blended price views