Release Notes: v0.14.0¶
Release Date: 2026-03-16
Highlights¶
- Three New Chinese LLM Providers: Added support for GLM (Zhipu AI), Kimi (Moonshot AI), and Qwen (Alibaba Cloud)
- Provider Conformance Testing: New
provider/providertestpackage for standardized testing across all providers
New Providers¶
GLM (Zhipu AI BigModel)¶
Zhipu AI's GLM models are among China's most advanced LLMs, featuring MoE architecture and advanced thinking capabilities.
client, err := omnillm.NewClient(omnillm.ClientConfig{
Providers: []omnillm.ProviderConfig{
{Provider: omnillm.ProviderNameGLM, APIKey: os.Getenv("GLM_API_KEY")},
},
})
Available Models:
- GLM-5 series: Flagship 744B MoE model with forced thinking mode
- GLM-4.7 series: Premium models with Interleaved Thinking
- GLM-4.5 series: Unified reasoning, coding, and agent capabilities
Kimi (Moonshot AI)¶
Moonshot AI's Kimi models excel at long-context understanding and multimodal tasks.
client, err := omnillm.NewClient(omnillm.ClientConfig{
Providers: []omnillm.ProviderConfig{
{Provider: omnillm.ProviderNameKimi, APIKey: os.Getenv("KIMI_API_KEY")},
},
})
Available Models:
- Kimi K2.5: Most intelligent multimodal model with 256k context
- Kimi K2 series: MoE foundation models with 1T total params
- Moonshot V1 series: General text generation with 8k/32k/128k context
Qwen (Alibaba Cloud)¶
Alibaba Cloud's Qwen (Tongyi Qianwen) models offer wide availability including US regions.
client, err := omnillm.NewClient(omnillm.ClientConfig{
Providers: []omnillm.ProviderConfig{
{Provider: omnillm.ProviderNameQwen, APIKey: os.Getenv("QWEN_API_KEY")},
},
})
Available Models:
- Qwen3 Max: Latest flagship with thinking capability
- QwQ series: Deep reasoning models with extended chain-of-thought
- Qwen2.5 series: Open-source instruction-following models
Provider Conformance Testing¶
New provider/providertest package provides standardized testing for all providers:
import "github.com/plexusone/omnillm/provider/providertest"
func TestConformance(t *testing.T) {
p := NewProvider(apiKey, "", nil)
providertest.RunAll(t, providertest.Config{
Provider: p,
SkipIntegration: apiKey == "",
TestModel: "your-model",
})
}
Test Categories:
- Unit tests (no API required): Interface compliance, streaming contracts
- Integration tests (API required): Real API calls with actual responses
Bug Fixes¶
- Streaming buffer limits: GLM and Kimi providers now handle large SSE payloads (64KB initial, 1MB max buffer)
- Empty choices handling: All new providers return explicit errors when API returns empty choices
- Lint compliance: Fixed
unparamissues in test files and gofmt formatting
Installation¶
Environment Variables¶
New environment variables for the Chinese LLM providers:
| Variable | Description |
|---|---|
GLM_API_KEY |
Zhipu AI BigModel API key |
KIMI_API_KEY |
Moonshot AI API key |
QWEN_API_KEY |
Alibaba Cloud DashScope API key |