Table of Contents

Choose controls the selected model supports

Claude Sonnet 5.5: Requires Mythosia.AI 8.2.0 / Abstractions 4.2.0. Configuration and migration

ThinkingToggle controls up-front thinking; None is not a promise of zero thinking blocks. MaxOutputTokens is 128000. FastSpeed, AsyncFunctionCalling and Steering are unsupported.

Grok 4.7: Requires Mythosia.AI 8.1.0 / Abstractions 4.1.0. model selection, reasoning and processing speed

GPT-6.1 Sol: Requires Mythosia.AI 8.2.0 / Abstractions 4.2.0. Selection and migration

GPT-6 Sol/Luna: Requires Mythosia.AI 8.1.0 / Abstractions 4.1.0. model selection and requirements

For Claude Opus 5.5, capabilities expose Low through Max, including XHigh; None and Minimal are unsupported. ThinkingToggle is unsupported and MaxOutputTokens is 128000. Hidden display does not mean thinking is disabled. Requires Mythosia.AI 8.1.0 / Abstractions 4.1.0.

A chat screen should offer reasoning, search, tools, or image controls that match the selected connection. Keeping model-name lists in every app duplicates the library’s rules and drifts when a provider, protocol, or deployment changes. Capability snapshots let the app and execution validation use the same model definitions.

This API belongs to Mythosia.AI 8.0.0 / Abstractions 4.0.0. The snapshots are immutable local descriptions of library support, not live account or server probes. Types are in Mythosia.AI.Models.Capabilities.

For requests where waiting time matters, use processing speed: WithSpeed keeps the model and reasoning effort, while Processing reports what the provider actually applied. Fast is a paid option on supported combinations.

Before / After

Before: the application maintains its own model lists. The lists below represent application code, not library APIs.

bool showReasoning = mySupportedReasoningModels.Contains(service.Model);
bool showSearch = mySupportedSearchModels.Contains(service.Model);

After: inspect the configured request, then choose supported options. The final completion call below performs the model request; inspecting capabilities does not.

using Mythosia.AI.Models;
using Mythosia.AI.Models.Capabilities;

var request = service.CreateRequest("Explain the documents.");
AIModelCapabilities capabilities = request.GetCapabilities();

if (capabilities.GetReasoningSupport(ReasoningLevel.High)
    == CapabilitySupport.Supported)
{
    request = request.WithReasoning(ReasoningLevel.High);
}

string answer = await request.GetCompletionAsync();

CapabilitySupport distinguishes Supported, Unsupported, and Unknown. Unknown means the library lacks enough information, such as a custom deployment or server-selected model; it does not mean unsupported. The example enables extra reasoning only when known to be supported. For unknown support, choose an application policy such as retaining defaults or allowing an attempted request.

request.GetCapabilities() reads the builder’s captured model, provider options and profile. service.GetCapabilities() inspects service defaults without consuming pending next-call options. Neither sends HTTP, invokes context callbacks or execution validators, changes history, nor starts work. Returned lists are read-only snapshots. The service query also peeks at pending next-call feature settings; it leaves them available for the actual request. Inspection does not serialize function defaults or hosted-tool parameters, and does not run execution profile preparation or token-budget reservations.

Capabilities describe what the connection can support, not which options you have enabled. Provider, API protocol and mode matter as well as the model name. The explicit wire model reflects provider overrides and Qwen/Ollama ID translation; when no single model is selected, model identity can be null. The sample Chat UI refreshes controls from the active connection and its current settings, including registered tools, rather than relying only on the model catalogue. Sampling support can change with reasoning mode or tool availability; query again after those settings change. Unknown support remains visibly distinct from unsupported support.

API Meaning
Reasoning, ReasoningLevels, GetReasoningSupport(...) Support and levels for common WithReasoning settings.
NativeReasoning, NativeReasoningLevels, ThinkingBudgetPresets, ThinkingToggle Provider-native reasoning controls and suggested budget choices.
Streaming, FunctionCalling, AsyncFunctionCalling, Steering Streaming, tools, provider-native asynchronous tools and mid-run instructions.
WebSearch, FileSearch, ReasoningCachePreservation, ImageInput, StructuredOutput Hosted search, cache-preserving reasoning changes, input images and structured output.
Temperature, TopP, FrequencyPenalty, PresencePenalty, MaxOutputTokens Supported sampling controls and a nullable known output-token limit.
StandardSpeed, FastSpeed, GetSpeedSupport(...) Mythosia.AI 8.1.0 / Abstractions 4.1.0: Supported/Unsupported/Unknown processing modes; account access is checked separately.
Provider, Model Provider and selected wire-model identity; either may be unknown.

Common and native reasoning are separate. ReasoningLevels describes common WithReasoning values; NativeReasoningLevels describes the provider’s own controls. ThinkingBudgetPresets supplies useful UI choices, not every valid budget or an exhaustive numeric range. AsyncFunctionCalling means native asynchronous tool execution, not merely that a local handler returns Task or runs in parallel. StructuredOutput covers the common typed-output API, including prompt-and-repair fallback; it does not guarantee native constrained decoding. Both reasoning-level lists use ReasoningLevel; budget presets are integer values.

A capability snapshot does not grant account access, establish live server readiness, or make malformed option combinations valid. Execution keeps its existing validation and errors. Check run.CanSteer on the actual running session before steering; a supported model capability alone does not mean the run is still active.

Inspect image generation separately

Image generation uses an independent model selection. Call service.GetImageCapabilities(imageModel) for a specific image model, or omit it for that provider’s image default. A chat request builder does not select an image-generation model. Use Generation, Editing and Mask to decide which image actions to present.

using Mythosia.AI.Models.Capabilities;

ImageModelCapabilities capabilities = service.GetImageCapabilities();
bool showEditing = capabilities.Editing == CapabilitySupport.Supported;
bool showMask = capabilities.Mask == CapabilitySupport.Supported;

foreach (var quality in capabilities.Qualities)
    Console.WriteLine(quality);

int? maximumImages = capabilities.MaxImages;

Qualities, Backgrounds, OutputFormats, SizeKinds, Resolutions and AspectRatios are typed, read-only option lists. MaxImages and MaxInputImages are nullable known limits. A listed option is not a promise that every combination is valid: existing size, format, quality, mask and model validation still applies. Custom or unknown image models retain unknown support instead of being labelled unsupported.

For Google, Resolutions and AspectRatios depend on the selected image model and also govern generation/editing validation. See the model-specific table, including the conservative Flash-Lite 1K policy. Unsupported explicit choices fail before HTTP; custom unknown models retain Unknown capabilities and provider-wide option validation.

For a custom AIService, override the protected ResolveRequestCapabilities() hook when the provider can supply reliable definitions. The default is AIModelCapabilities.Unknown. Adding a catalogue entry must not turn an unknown deployment into an unsupported one. IAIService consumers do not gain mandatory capability members; these query methods belong to AIService and its request builder.

If a custom provider profile changes native mode flags, override ApplyCapabilityRequestProfile(AIRequestProfile) and use SetExecutionSetting(...) only for the flags needed by the resolver. The default hook does nothing. Common profile overrides are already captured by the builder; inspection never calls ApplyRequestProfile or ApplyProviderSpecificRequestProfile. Keep this hook free of validation, callbacks, serialization, budget reservation and changes to service or caller-owned state. Its temporary settings are restored after inspection, including when an override throws.

Request settings · Provider and image options · Run control

Edit this page