I found One Million Tokens. Its rough scale:
1M tokens
~ 750K words
~ 3,000 pages
~ 83 hours of conversation
~ 75K lines of code
🧠 The more interesting part is the timeline.
It starts with GPT-3 at 2,048 tokens in 2020, then walks through ChatGPT 4K, GPT-4 32K, Claude 100K, Gemini 1M, and the later multi-million-token era.
The visual change is kind of absurd when you see all the pages stacked together.
One caveat: maximum context is not the same as perfect memory/retrieval. A model accepting 1M tokens can still fail to use information buried inside that window.
The line-of-code/page conversions are approximations too. I am curious how people here design around 1M+ context in practice.
Do you actually feed giant source sets directly, or still prefer retrieval + smaller focused contexts for cost/attention reasons?
In what world is a production codebase 75k lines. A single project at my last company was 1.4 million lines of code. Every single one of our microservices had more lines of code than that. And just my team was responsible for 50+ microservices. And thirteen tokens per line is hilariously low.
Holding a week of chat in your head is nothing when 90% of it is stuff that is incorrect or discussed and minds changed. This is exactly why my boss keeps saying conflicting things, the LLM doesn’t know what is true or not. It just regurgitates the random thing from context.
LLM post about LLMs
lowend llm inference service prices ( 2 random providers. not extensive pricing research)
nex-n2-mini costs 0.025 / 0.10 ( per Mtok USD )
gemma 4 E2B 5B 0.01 / 0.03I can also show you what 1 Million tokens looks like, behold:
.
.
.
(this space left intentionally blank)
Rookie numbers.
