Most commercial LLM creators aren't that transparent about parameter count.
I'm wondering which of the top 2 LLMs now (March 2025 - at least according lmarena.ai and presumably the two largest ones right now) uses more parameters.
I suspect both are on the order of 4T (give or take a factor of 2) but I don't know for sure.
Resolves when it is generally well known which one is a larger model.
If unknown by 2030, resolves either N/A or PROB if there's strong (but not conclusive) evidence on the parameter counts of GPT-4.5 and Grok 3.
Context (based on what I know/heard)
GPT 3/3.5: 175B
GPT 3.5 Turbo: 20B
GPT 4: 1.8T
GPT 4-Turbo: less? ~200B?
GPT 4o: 200B?
GPT 4o-mini: 20B?
GPT 4.5:?
Grok 1: 314B
Grok 2:?
Grok 3:?
Deepseek v3 + r1: 671B
Here's the AI Generated Description (not sure if it's trustworthy)
Which LLM Has More Parameters: GPT-4.5 vs Grok 3
Background
Large Language Models (LLMs) have been growing in size and capability, with parameter count often serving as one metric of model scale. Based on available information:
GPT-4.5 has been confirmed to have approximately 12.8 trillion parameters (12.8T)
Grok 3's exact parameter count has not been officially disclosed by xAI
There are unconfirmed speculations that Grok 3 may have hundreds of billions of parameters, potentially around 400 billion for future versions like Grok 3.5
For context on previous models:
GPT-4 was estimated to have around 1.8T parameters
Grok 1 had 314B parameters
Resolution Criteria
This market will resolve to whichever model (GPT-4.5 or Grok 3) has the higher parameter count once this information becomes generally well-known and accepted.
If official information from OpenAI and xAI confirms parameter counts, the market will resolve based on that data
If by 2030 the exact parameter counts remain unknown, but there is strong evidence pointing to one model being larger, the market may resolve probabilistically
If by 2030 there is insufficient evidence to determine which model has more parameters, the market will resolve N/A
Considerations
Parameter count is just one metric of model capability and doesn't necessarily correlate directly with performance. Models can differ in architecture, training data, and optimization techniques that affect their capabilities beyond raw parameter count.
Based on currently available information, GPT-4.5's 12.8T parameters would significantly exceed Grok 3's presumed hundreds of billions of parameters, but official confirmation from xAI regarding Grok 3's parameter count is still needed for definitive resolution.