Model configurations

NeuralPlus keeps model names as configuration data. The SDK therefore has one client class per provider protocol, not one C++ class per marketed model.

For common choices, neuralplus/models.hpp provides ready-to-use typed configurations:

#include <neuralplus/neuralplus.hpp>
 
#include <utility>
 
auto config = neuralplus::models::openai::gpt_5_6_terra();
auto client = neuralplus::make_client(std::move(config));

The result is an ordinary OpenAIConfig, AnthropicConfig, or GeminiConfig. Applications can still set credentials, URLs, headers, capabilities, and provider options before passing it to make_client.

Included configurations

The catalog was checked against provider documentation on July 23, 2026. Provider availability, aliases, limits, and pricing can change independently of an SDK release, so check the linked source before selecting a production model.

Provider	C++ configuration	Model ID
OpenAI	`models::openai::gpt_5_6_sol()`	`gpt-5.6-sol`
OpenAI	`models::openai::gpt_5_6_terra()`	`gpt-5.6-terra`
OpenAI	`models::openai::gpt_5_6_luna()`	`gpt-5.6-luna`
Anthropic	`models::anthropic::claude_fable_5()`	`claude-fable-5`
Anthropic	`models::anthropic::claude_opus_4_8()`	`claude-opus-4-8`
Anthropic	`models::anthropic::claude_sonnet_5()`	`claude-sonnet-5`
Anthropic	`models::anthropic::claude_haiku_4_5()`	`claude-haiku-4-5-20251001`
Gemini	`models::gemini::gemini_3_6_flash()`	`gemini-3.6-flash`
Gemini	`models::gemini::gemini_3_5_flash()`	`gemini-3.5-flash`
Gemini	`models::gemini::gemini_3_5_flash_lite()`	`gemini-3.5-flash-lite`
Gemini	`models::gemini::gemini_2_5_pro()`	`gemini-2.5-pro`

Sources:

The catalog describes capabilities used by NeuralPlus before a request is sent. It does not add a capability that the provider or account does not have.

Use a custom or newly released model

No catalog update is required to use another model:

neuralplus::OpenAIConfig config{"gpt-new-or-fine-tuned-id"};
config.model.display_name = "My application model";
config.model.capabilities.image_input = true;
config.model.capabilities.file_input = true;
config.model.capabilities.tools = true;
config.model.provider_options["reasoning"] = {{"effort", "medium"}};
 
auto client = neuralplus::make_client(std::move(config));

The same pattern works with AnthropicConfig and GeminiConfig. OpenAICompatibleConfig always uses an application-supplied server URL and model ID because compatible servers do not share one authoritative catalog.

Use typed GenerateOptions for portable controls. Use ModelDescriptor::provider_options for a model-wide provider field and GenerateOptions::provider_options for a one-call override. A new wire protocol belongs in its provider adapter; a new marketed model normally needs configuration only.