Skip to content

model-providers.md

Latest commit

 

History

History
240 lines (176 loc) · 6.13 KB

model-providers.md

File metadata and controls

240 lines (176 loc) · 6.13 KB
Raw

Model Providers

agentscope-go supports 9 LLM providers, all implementing the ChatModel interface with Chat, ChatStream, and CountTokens.

Configuration Pattern

Every provider follows the same pattern:

cm, err := model.NewXxxChatModel(model.XxxConfig{
    APIKey: "...",
    Model:  "model-name",
    // Optional:
    ClientOptions: &model.ClientOptions{
        Timeout:        120 * time.Second,
        DefaultHeaders: map[string]string{"X-Custom": "value"},
        Transport:      customTransport, // for proxy
    },
})
// Note: Anthropic and OpenAI Response take pointer configs (*AnthropicConfig, *OpenAIResponseConfig)

Providers

OpenAI

cm, _ := model.NewOpenAIChatModel(model.OpenAIConfig{
    APIKey: os.Getenv("OPENAI_API_KEY"),
    Model:  "gpt-4o",
})

Models: gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-5.4, gpt-5.5, gpt-audio-mini, o3, o4-mini

OpenAI Responses API

cm, _ := model.NewOpenAIResponseModel(&model.OpenAIResponseConfig{
    APIKey:          os.Getenv("OPENAI_API_KEY"),
    Model:           "gpt-4.1",
    ReasoningEffort: "high",
})

Uses the Responses API (/v1/responses) instead of Chat Completions. Default timeout is 5 minutes for long reasoning tasks.

Anthropic

cm, _ := model.NewAnthropicChatModel(&model.AnthropicConfig{
    APIKey:          os.Getenv("ANTHROPIC_API_KEY"),
    Model:           "claude-sonnet-4-20250514",
    MaxOutputTokens: 4096,
})

Models: claude-opus-4-5, claude-opus-4-6, claude-opus-4-7, claude-opus-4-8, claude-sonnet-4-5, claude-sonnet-4-6, claude-haiku-4-5

Extended thinking: model.WithThinking(true, 10000) enables thinking with a 10K token budget.

DashScope (Alibaba Qwen)

cm, _ := model.NewDashScopeChatModel(model.DashScopeConfig{
    APIKey:  os.Getenv("DASHSCOPE_API_KEY"),
    BaseURL: os.Getenv("DASHSCOPE_BASE_URL"), // optional override
    Model:   "qwen-plus",
})

Models: qwen-long, qwen-plus, qwen3.5-omni-plus, qwen3.5-plus, qwen3.6-max-preview, qwen3.6-plus, qwen3.7-max, qwen3.7-plus

GLM models via DashScope: glm-5.2 (131K context, 8K output)

DeepSeek

cm, _ := model.NewDeepSeekChatModel(model.DeepSeekConfig{
    APIKey: os.Getenv("DEEPSEEK_API_KEY"),
    Model:  "deepseek-chat",
})

Models: deepseek-chat, deepseek-reasoner, deepseek-v4-flash, deepseek-v4-pro

Google Gemini

cm, _ := model.NewGeminiChatModel(model.GeminiConfig{
    APIKey: os.Getenv("GEMINI_API_KEY"),
    Model:  "gemini-2.5-flash",
})

Models: gemini-2.5-flash, gemini-2.5-pro, gemini-3-flash-preview, gemini-3.1-pro-preview

Ollama (Local)

cm, _ := model.NewOllamaChatModel(model.OllamaConfig{
    BaseURL: "http://localhost:11434", // default
    Model:   "llama4",
})

No API key required. Models: deepseek-r1-14b, llama4, qwen3-14b, qwen3.5-9b

Moonshot (Kimi)

cm, _ := model.NewMoonshotChatModel(model.MoonshotConfig{
    APIKey: os.Getenv("MOONSHOT_API_KEY"),
    Model:  "kimi-k2.6",
})

Models: kimi-k2.5, kimi-k2.6, kimi-k3, moonshot-v1-8k, moonshot-v1-32k, moonshot-v1-128k

Kimi K3 — Latest Kimi model with 256K context, 64K output. Supports text, vision (JPEG/PNG/GIF/WebP), video (MP4), and extended thinking.

xAI (Grok)

cm, _ := model.NewXAIChatModel(model.XAIConfig{
    APIKey: os.Getenv("XAI_API_KEY"),
    Model:  "grok-3",
})

Models: grok-3, grok-3-fast, grok-3-mini, grok-4.3

Text-to-Speech (TTS)

agentscope-go includes TTS models for audio synthesis, usable standalone or via TTSMiddleware.

DashScope TTS

Standard and realtime (CosyVoice) TTS:

ttsModel, _ := tts.NewDashScopeTTSModel(tts.DashScopeConfig{
    APIKey: os.Getenv("DASHSCOPE_API_KEY"),
    Model:  "cosyvoice-v3-flash",
    Voice:  "longxiaochun",
})
audio, _ := ttsModel.Synthesize(ctx, "Hello world")

Models: cosyvoice-v1, cosyvoice-v2, cosyvoice-v3-flash, cosyvoice-v3-plus, qwen3-tts-flash, qwen3-tts-flash-realtime

OpenAI TTS

ttsModel, _ := tts.NewOpenAITTSModel(&tts.OpenAITTSConfig{
    APIKey: os.Getenv("OPENAI_API_KEY"),
    Model:  "tts-1-hd",
    Voice:  "alloy",
})

Models: tts-1, tts-1-hd

Gemini TTS

Uses the Gemini generateContent API with audio response modality:

ttsModel, _ := tts.NewGeminiTTSModel(tts.GeminiTTSConfig{
    APIKey:       os.Getenv("GEMINI_API_KEY"),
    Model:        "gemini-2.5-flash-preview-tts",
    Voice:        "Kore",
    SpeakingRate: 1.0,
})
audio, _ := ttsModel.Synthesize(ctx, "Hello from Gemini TTS")

Models: gemini-2.5-flash-preview-tts

Common Call Options

All providers support the same CallOption set:

resp, _ := cm.Chat(ctx, msgs,
    model.WithTemperature(0.7),
    model.WithMaxTokens(1024),
    model.WithTopP(0.9),
    model.WithTools(toolSchemas),
    model.WithToolChoice(&model.ToolChoice{Mode: "auto"}),
    model.WithThinking(true, 5000),
    model.WithReasoningEffort("high"),
    model.WithVoice("alloy"),
    model.WithRetries(3, time.Second),
)

Fallback Model

Automatic failover from a primary to a backup provider:

cm := model.NewFallbackChatModel(primaryModel, fallbackModel)

Structured Output

Force any model to produce JSON matching a schema:

result, _ := model.GenerateStructuredOutput(ctx, cm, msgs, jsonSchema)

Model Cards

Query bundled model metadata:

card, _ := model.GetModelCard("claude-sonnet-4-6")
// card.Name, card.Provider, card.ContextSize, card.OutputSize

models := model.ListModels()          // all 54 chat model cards
models = model.ListModels("openai")   // filter by provider

54 chat/embedding model cards + 9 TTS model cards are bundled and accessible at runtime via //go:embed.

Model Card Summary

Provider Chat Models Embedding Models TTS Models
OpenAI 10 2 2
Anthropic 7 — —
DashScope 9 (incl. GLM-5.2) 2 6
DeepSeek 4 — —
Gemini 4 2 1
Moonshot 6 (incl. Kimi K3) — —
Ollama 4 — —
xAI 4 — —
Total 48 6 9