LLM price indexWhat LLMs cost: November 2026
In November 2026, the cheapest frontier model (Gemini 3.1 Pro) costs $1,406 a month for a standard 500-user workload. The cheapest balanced model (Gemini 3.8 Flash) costs $479 and the cheapest efficient model (Claude Haiku 5.5) $64, so frontier costs 22.0 times as much as efficient.
By Ramesh Emandi, Co-Founder and CTO ·
Prices for ten models
| Model | Input | Output | Cached input | Batch in / out | Workload cost |
|---|---|---|---|---|---|
| Claude Opus 5.5 Anthropic · frontier | $4.00 | $20.00 | $0.20 | $2.00 / $10.00 | $2,552 |
| GPT-6 Astra OpenAI · frontier | $10.00 | $50.00 | $1.00 | $5.00 / $25.00 | $6,380 |
| Gemini 3.1 Pro Google · frontier · cheapest in tier | $2.00 | $12.00 | $0.20 | $1.00 / $6.00 | $1,406 |
| Claude Sonnet 5.5 Anthropic · balanced | $2.00 | $10.00 | $0.10 | $1.00 / $5.00 | $1,276 |
| GPT-6.1 Sol OpenAI · balanced | $2.00 | $10.00 | $0.10 | $1.00 / $5.00 | $1,276 |
| Gemini 3.8 Flash Google · balanced · cheapest in tier | $0.75 | $3.75 | $0.075 | $0.375 / $1.875 | $479 |
| Claude Haiku 5.5 Anthropic · efficient · cheapest in tier | $0.10 | $0.50 | $0.01 | $0.05 / $0.25 | $64 |
| GPT-6.1 Terra OpenAI · efficient | $1.00 | $5.00 | $0.05 | $0.50 / $2.50 | $638 |
| Gemini 3.5 Flash-Lite Google · efficient | $0.30 | $2.50 | $0.03 | $0.15 / $1.25 | $256 |
| DeepSeek V4 Flash DeepSeek · efficient · open weights | $0.30 | $1.20 | $0.006 | n/a | $172 |
The workload cost prices 500 users, each running 10 chats (1,500 tokens in, 400 out), 2 documents (6,000 in, 800 out) and 1 image (1,500 in, 300 out) a day over 22 working days, at list prices with no caching, batching or routing. Change the assumptions in the AI cost calculator.
What changed since last month
This is the first edition of the index, so there is no earlier month to compare against. From the next edition this section lists every model whose input, output, cached or batch price has moved, with the change in workload cost.
Baseline set on 6 November 2026: 10 models, no changes recorded yet.
Methodology and source
Prices are list prices in US dollars per million tokens, taken from the LiteLLM model price list, a community-maintained record published under the MIT licence. We read it on 6 November 2026. Cached input is the price for tokens served from the provider’s prompt cache. Batch is the asynchronous batch API price. Where a provider publishes no such price, the table shows n/a.
We track one model per tier for each vendor: frontier for the most capable, balanced for everyday work, and efficient for high-volume, low-cost tasks. Tiers are our grouping, not the vendors’. Prices are a secondary-source record and exclude volume discounts, negotiated rates, regional uplifts and tax. Confirm on each provider’s own pricing page before budgeting.
To cut these figures further, read our guide on how to reduce LLM token usage and AI running costs, or run your own numbers in the AI cost calculator.
Archive
- November 2026 (current, read 6 November 2026)
A new edition is published each month. Earlier editions stay on this page.
LLM pricing, answered.
What is the Quantana LLM price index?
A monthly, dated record of what ten widely used large language models cost per million tokens, across Anthropic, OpenAI, Google and DeepSeek. Each edition also prices the same standard workload on every model so the figures can be compared directly.
How is the monthly workload cost calculated?
The same assumptions apply to every model: 500 users, 10 chats, 2 documents and 1 image a day each, over 22 working days, at list prices with no caching, batching or routing. It uses the logic in our AI cost calculator.
Where do the prices come from?
From the LiteLLM community price list on GitHub, released under the MIT licence, read on the date shown on each edition. It is a secondary source, so confirm on the provider’s own pricing page before you commit to a budget.
Why do cached and batch prices matter?
Cached input and batch processing are often a fraction of the standard price. Workloads that repeat context or can wait for results can cost far less than the list price suggests. Our guide to reducing AI costs covers when each applies.
Talk to us about reducing your AI costs
List prices are only the starting point. Caching, batching, routing and the right model for each task usually matter more. Talk to Quantana and we can review where your tokens go and what a better setup would cost.