How fast are LLM prices falling?
Data as of 2 Nov 2026 · last updated · by BenchLeader
The cheapest model scoring 60 or more on the BenchLeader Index costs $1.50 per 1M tokens blended (Gemini 3.7 Flash); in April 2025, when the level was first reached, the cheapest was o3 at $3.50, 2.3× as much, at today's list prices. Each level followed is cheaper than when it was first reached, by 2.3× to 279×.
Key statistics
Each statistic has its own link (#) to cite it directly.
- The cheapest model scoring 50 or more on the BenchLeader Index costs $0.094 per 1M tokens blended (Qwen3.5 Flash); in December 2024, when the level was first reached, the cheapest was o1 at $26.25, 279× as much (BenchLeader, 2 Nov 2026). #
- The cheapest model scoring 55 or more on the BenchLeader Index costs $0.15 per 1M tokens blended (Step 3.5 Flash); in April 2025, when the level was first reached, the cheapest was o4-mini at $1.93, 13× as much (BenchLeader, 2 Nov 2026). #
- The cheapest model scoring 60 or more on the BenchLeader Index costs $1.50 per 1M tokens blended (Gemini 3.7 Flash); in April 2025, when the level was first reached, the cheapest was o3 at $3.50, 2.3× as much (BenchLeader, 2 Nov 2026). #
- The cheapest model scoring 65 or more on the BenchLeader Index costs $2.00 per 1M tokens blended (GPT-6.1 Terra); in June 2026, when the level was first reached, the cheapest was Claude Fable 5 at $20.00, 10× as much (BenchLeader, 2 Nov 2026). #
- Every quality level BenchLeader follows is cheaper than when it was first reached: from 2.3× to 279× cheaper, comparing today's list prices of the cheapest model then and now (BenchLeader, 2 Nov 2026). #
The cheapest model at each level, as it changed
For each level, the month a new model became the cheapest way to reach it, newest first, at today's list price blended three input tokens to one output.
Index 50 or more279× cheaper than at first
| From | Cheapest model | $ / 1M blended |
|---|---|---|
| February 2026 | Qwen3.5 Flash | $0.094 |
| November 2025 | Grok 4.1 Fast | $0.275 |
| September 2025 | Grok 4 Fast | $0.275 |
| July 2025 | Qwen3 235B A22B 2507 | $0.368 |
| April 2025 | Qwen3 235B A22B | $0.502 |
| March 2025 | Gemini 2.5 Pro | $3.44 |
| February 2025 | Grok 3 | $6.00 |
| December 2024 | o1 | $26.25 |
Index 55 or more13× cheaper than at first
| From | Cheapest model | $ / 1M blended |
|---|---|---|
| April 2026 | Step 3.5 Flash | $0.15 |
| November 2025 | Grok 4.1 Fast | $0.275 |
| August 2025 | GPT-5 mini | $0.688 |
| April 2025 | o4-mini | $1.93 |
Index 60 or more2.3× cheaper than at first
| From | Cheapest model | $ / 1M blended |
|---|---|---|
| August 2026 | Gemini 3.7 Flash | $1.50 |
| July 2026 | Gemini 3.6 Flash | $1.50 |
| February 2026 | GLM-5 | $1.55 |
| April 2025 | o3 | $3.50 |
Index 65 or more10× cheaper than at first
| From | Cheapest model | $ / 1M blended |
|---|---|---|
| October 2026 | GPT-6.1 Terra | $2.00 |
| September 2026 | GPT-6.1 Sol | $4.00 |
| July 2026 | GPT-5.6 Sol | $8.00 |
| June 2026 | Claude Fable 5 | $20.00 |
How these numbers are made
Read back from release dates: for each month, the models released by its end that score at or above each level on today's BenchLeader Index, at today's list price. Prices are today's rather than the launch price, models listed at $0 are left out, and so are models with fewer than 10 independent results. Models BenchLeader no longer tracks are not in it, so an early month can look dearer than it was when a cheaper model has since been retired.
Today's prices alone are on LLM API price statistics. Over a longer run, Epoch AI estimates the cost of reaching a given level of performance has fallen about 47% per quarter since 2023 (The plunging price of thought).
Cite these statistics
List prices are each provider's published rates, read via Artificial Analysis, LiteLLM price table, OpenRouter, models.dev. They are shown here with attribution and are not offered for reuse; BenchLeader's own figures on this page are the Index scores. Cite the page below.
Citation
BenchLeader, “LLM price trends”, data as of 2 Nov 2026, https://www.benchleader.com/stats/llm-price-trends.Questions people ask
- How fast are LLM prices falling?
- The cheapest model scoring 60 or more on the BenchLeader Index costs $1.50 per 1M tokens blended (Gemini 3.7 Flash); in April 2025, when the level was first reached, the cheapest was o3 at $3.50, 2.3× as much, at today's list prices. Each level followed is cheaper than when it was first reached, by 2.3× to 279×. Over the longer run, Epoch AI estimates the cost of reaching a given level of performance has fallen about 47% per quarter since 2023 (“The plunging price of thought”).
- How much does a top AI model cost per million tokens?
- 50 or more: $0.094 (Qwen3.5 Flash); 55 or more: $0.15 (Step 3.5 Flash); 60 or more: $1.50 (Gemini 3.7 Flash); 65 or more: $2.00 (GPT-6.1 Terra). Blended at three input tokens to one output token.
- How is this measured?
- For each month, the cheapest model among those released by the month's end that scores at or above each level on today's BenchLeader Index, at today's list price blended 3:1 input to output. Models listed at $0 are left out, and so are models with fewer than 10 independent results. Prices are today's, not the launch price, and models BenchLeader no longer tracks are not in it, so earlier months can look dearer or cheaper than they were.
Related: LLM prices today, LLM cost calculator, best value models, the AI race, AI trends and statistics.