Max Tokens are Wrong for new Claude 4 models

[harrisreynolds]

Currently set to 4096 for max output but they are actually 32 to 64K

This is a total blocker for me... I feel like the current model registry is a nice idea but just hard to keep up to date.

We need a way to override it.

See details below:

#<RubyLLM::ModelInfo:0x000000012b4f7280 @id="claude-opus-4-20250514", @created_at=2025-05-22 00:00:00 UTC, @display_name="Claude Opus 4", @provider="anthropic", @context_window=200000, @max_tokens=4096, @type="chat", @family=:claude2, @supports_vision=true, @supports_functions=false, @supports_json_mode=false, @input_price_per_million=3.0, @output_price_per_million=15.0, @metadata={}>
#<RubyLLM::ModelInfo:0x000000012b4f70a0 @id="claude-sonnet-4-20250514", @created_at=2025-05-22 00:00:00 UTC, @display_name="Claude Sonnet 4", @provider="anthropic", @context_window=200000, @max_tokens=4096, @type="chat", @family=:claude2, @supports_vision=true, @supports_functions=false, @supports_json_mode=false, @input_price_per_million=3.0, @output_price_per_million=15.0, @metadata={}>

[tpaulshippy]

I recommend forking or monkey patching to override for now.

[crmne]

Hi @harrisreynolds if there are any problems with the capabilities please file a bug to Parsera https://github.com/parsera-labs/api-llm-specs/issues

It also looks like the problem is solved. Closing!