Resolution criteria

• Each month option resolves YES if, during that calendar month (UTC), at least one of Google, OpenAI, or Anthropic makes a new large language model (LLM) publicly available beyond a private waitlist in any of: their consumer apps (Gemini/ChatGPT/Claude) or developer platforms/APIs (OpenAI API, Google AI Studio/Vertex AI, Anthropic API/AWS Bedrock/Vertex AI). “Publicly available” includes GA or an open preview accessible without individual approval. If availability and announcement fall in different months, use the first month of broad availability.

• “New LLM” = a newly named model or sub‑variant with a distinct model identifier added to the company’s official model/catalog (e.g., Gemini 2.5 Flash‑Lite, Claude Sonnet 4, OpenAI o3/o4‑mini/GPT‑4.5). Routine snapshot updates, pricing/context changes, UI features, or pure renames do not count. Anonymous or cloaked models do not count - if a cloaked model such as Horizon Alpha were released in one month and a company publicly claims it in the next, this market would count that model as released in the second month.

• Specialized coding models such as OpenAI's codex models, image generation models, or other special purpose models do not count.

• In the case of ambiguity the market will err on resolving no. In case of ambiguity about whether there is ambiguity I will use my judgement.

• Evidence must be from the company’s official pages (any one suffices):

  • OpenAI blog or Help Center release notes. (openai.com)

  • Google DeepMind/Google AI blog or Gemini models docs. (blog.google)

  • Anthropic news/docs model lists. (anthropic.com)

• A month resolves NO if none of the three release a qualifying new LLM in that month. If multiple releases occur in a month, the option still resolves YES once. If a model is briefly released and pulled, it counts if availability lasted ≥24 hours and was documented by the above sources.

Considerations

• Companies often “announce” before GA; this market uses the first month of broad availability. Experimental/preview models count only if openly accessible (no private tester gate).

• Releases via partner platforms (AWS Bedrock, Vertex AI) count if they debut there first.

• Minor updates to existing models (e.g., improved 2.5 Flash without a new model name) do not qualify; new named variants like “2.5 Flash‑Lite” or “Opus 4.1” do.