The BenchLM Token Price Index
The BenchLM Token Price Index stands at 18.7 for September 2026 — frontier LLM token prices are 81.3% below their March 2023 base, at a median blended $7.00 per 1M tokens across 22 flagship models.
Published monthly · Base: March 2023 = 100 · Methodology v1.2
Frontier index · September 2026
18.7
March 2023 = 100 · lower is cheaper
- Median blended
- $7.00 / 1M
- Active constituents
- 22
- Blend
- 3:1 input:output
Index by tier, March 2023 – September 2026 (each tier's first month = 100)
| Sub-index | Value | MoM | YoY |
|---|---|---|---|
| Frontier | 18.7 | +16.9% | +101.1% |
| Mid-tier | 32.1 | -35.8% | -40.8% |
| Budget | 91.3 | -21.5% | +7.4% |
What moved since the last snapshot
- Claude Opus 5.5Frontier — entered the index at $8.00/1M blended
- GPT-5.6 SolFrontier — price cut from $11.25 to $8.00/1M blended (-28.9%)
- GPT-5.6 TerraMid-tier — price cut from $5.63 to $4.50/1M blended (-20%)
- GPT-5.6 LunaBudget — price cut from $2.25 to $0.45/1M blended (-80%)
- Gemini 3.6 FlashMid-tier — price cut from $3.00 to $1.50/1M blended (-50%)
- Claude Fable 5.1Frontier — entered the index at $20.00/1M blended
Current constituents (41)
| Model | Tier | Input $/1M | Output $/1M | Blended $/1M |
|---|---|---|---|---|
| DeepSeek R1 (DeepSeek) | Frontier | $0.55 | $2.19 | $0.96 |
| Grok 4.6 (xAI) | Frontier | $2 | $6 | $3.00 |
| 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 |
| Kimi K3 (Moonshot AI) | Frontier | $3 | $15 | $6.00 |
| Claude Opus 5.5 (Anthropic) | Frontier | $4 | $20 | $8.00 |
| GPT-5.6 Sol (OpenAI) | Frontier | $4 | $20 | $8.00 |
| Claude Opus 4.6 (Anthropic) | Frontier | $5 | $25 | $10.00 |
| Claude Opus 4.8 (Anthropic) | Frontier | $5 | $25 | $10.00 |
| Claude Opus 5 (Anthropic) | Frontier | $5 | $25 | $10.00 |
| GPT-5.5 (OpenAI) | Frontier | $5 | $30 | $11.25 |
| Claude Fable 5 (Anthropic) | Frontier | $10 | $50 | $20.00 |
| Claude Fable 5.1 (Anthropic) | Frontier | $10 | $50 | $20.00 |
| o1 (OpenAI) | Frontier | $15 | $60 | $26.25 |
| GPT-5.4 Pro (OpenAI) | Frontier | $30 | $180 | $67.50 |
| GPT-5.5 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 |
| Gemini 3.6 Flash (Google) | Mid-tier | $0.75 | $3.75 | $1.50 |
| 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 Sonnet 5 (Anthropic) | Mid-tier | $2 | $10 | $4.00 |
| GPT-5.6 Terra (OpenAI) | Mid-tier | $2 | $12 | $4.50 |
| 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 |
| GPT-5.6 Luna (OpenAI) | Budget | $0.2 | $1.2 | $0.45 |
| GPT-5.4 nano (OpenAI) | Budget | $0.2 | $1.25 | $0.46 |
| 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 for non-commercial use with that credit, under CC BY-NC 4.0. Commercial use needs a license.
BenchLM Token Price Index, September 2026: 18.7 (Mar 2023 = 100). BenchLM.ai, https://benchlm.ai/token-price-indexQuestions
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 18.7 means frontier token prices are 81.3% 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 18.7 for September 2026 — frontier LLM token prices are 81.3% below their March 2023 base, at a median blended $7.00 per 1M tokens across 22 flagship models. Prices are per blended 1M tokens across 22 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 issueJoin 2,000+ readers.
One email each week. Unsubscribe anytime.