Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start free brief

The BenchLM Token Price Index

Updated August 15, 2026 · Published monthly · Base: March 2023 = 100 · Methodology v1.0

30 confirmed releases in the last 30 daysStart free brief

Frontier index · August 2026

12

-3.2% MoM+36.4% YoY88% vs Mar 2023

The BenchLM Token Price Index stands at 12 for August 2026 — frontier LLM token prices are 88% below their March 2023 base, at a median blended $4.50 per 1M tokens across 13 flagship models.

Index by tier, March 2023August 2026 (each tier's first month = 100)

Sub-indexValueMoMYoY
Frontier12-3.2%+36.4%
Mid-tier32.1+6.6%-35.8%
Budget1000%+90.5%

What moved since the last snapshot

  • Claude 3.5 SonnetMid-tierentered the index at $6.00/1M blended
  • Gemini 1.5 ProFrontierentered the index at $2.19/1M blended

Current constituents (29)

ModelTierInput $/1MOutput $/1MBlended $/1M
DeepSeek R1 (DeepSeek)Frontier$0.55$2.19$0.96
Gemini 1.5 Pro (Google)Frontier$1.25$5$2.19
GPT-5.1 (OpenAI)Frontier$1.25$10$3.44
Gemini 2.5 Pro (Google)Frontier$1.25$10$3.44
o3 (OpenAI)Frontier$2$8$3.50
GPT-4o (OpenAI)Frontier$2.5$10$4.38
Gemini 3.1 Pro (Google)Frontier$2$12$4.50
GPT-5.2 (OpenAI)Frontier$1.75$14$4.81
GPT-5.4 (OpenAI)Frontier$2.5$15$5.63
Grok 4.1 (xAI)Frontier$3$15$6.00
Claude Opus 4.6 (Anthropic)Frontier$5$25$10.00
o1 (OpenAI)Frontier$15$60$26.25
GPT-5.4 Pro (OpenAI)Frontier$30$180$67.50
DeepSeek V3 (DeepSeek)Mid-tier$0.27$1.1$0.48
Mistral Large 3 (Mistral)Mid-tier$0.5$1.5$0.75
Gemini 3 Flash (Google)Mid-tier$0.5$3$1.13
GPT-5.4 mini (OpenAI)Mid-tier$0.75$4.5$1.69
o3-mini (OpenAI)Mid-tier$1.1$4.4$1.93
GPT-4.1 (OpenAI)Mid-tier$2$8$3.50
Claude 3.5 Sonnet (Anthropic)Mid-tier$3$15$6.00
Claude Sonnet 4.5 (Anthropic)Mid-tier$3$15$6.00
Claude Sonnet 4.6 (Anthropic)Mid-tier$3$15$6.00
GPT-4.1 nano (OpenAI)Budget$0.1$0.4$0.17
GPT-4o mini (OpenAI)Budget$0.15$0.6$0.26
Grok 3 Mini (xAI)Budget$0.3$0.5$0.35
Claude 3 Haiku (Anthropic)Budget$0.25$1.25$0.50
GPT-4.1 mini (OpenAI)Budget$0.4$1.6$0.70
Gemini 2.5 Flash (Google)Budget$0.3$2.5$0.85
Claude Haiku 4.5 (Anthropic)Budget$1$5$2.00

Blended = (3 × input + output) / 4. Constituent rules and tier assignments: methodology.

Cite the index

Cite as "the BenchLM Token Price Index" with a link to this page. The full monthly series is free to use with attribution:

BenchLM Token Price Index, August 2026: 12 (Mar 2023 = 100). BenchLM.ai, https://benchlm.ai/token-price-index

Frequently Asked Questions

What is the BenchLM Token Price Index?

The BenchLM Token Price Index tracks what 1M tokens costs at the frontier of AI, month by month. It is the median blended API price (3:1 input:output ratio) across the flagship models sold at each point in time, normalized so March 2023 (GPT-4 launch) = 100. A value of 12 means frontier token prices are 88% below that base. Mid-tier and budget sub-indexes track the same rule for workhorse and small models.

How much have LLM token prices dropped?

The BenchLM Token Price Index stands at 12 for August 2026 — frontier LLM token prices are 88% below their March 2023 base, at a median blended $4.50 per 1M tokens across 13 flagship models. Prices are per blended 1M tokens across 13 active frontier constituents.

How is the index calculated?

Each month, every active constituent gets a blended price of (3 × input + output) / 4 per 1M tokens. The index is the tier's median, normalized to its base month = 100. Constituents enter at launch and exit when superseded or no longer listed. The full rules, constituent list, and revision policy are published on the methodology page, and the index is recomputed from source data on every site build.

Can I cite or embed the index?

Yes. Cite it as "the BenchLM Token Price Index" with a link to benchlm.ai/token-price-index. The full series is downloadable at /data/price-index.json, and an embeddable chart is available at /embed/price-index.

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.