Provider & LLM Ecosystem Analysis — 2026-07¶

Date: 2026-07-03 Source: opencode source code (GitHub dev branch), provider documentation, market analysis

1. z.ai (Zhipu AI) — GLM Series¶

Current Models (July 2026)¶

API Endpoint¶

Base URL: https://api.z.ai/api/paas/v4
Auth: Bearer token (ZAI_API_KEY)
Compatible: OpenAI API format

Integration¶

{
  "provider": {
    "zai": {
      "options": {
        "baseURL": "https://api.z.ai/api/paas/v4",
        "apiKey": "{ZAI_API_KEY}"
      }
    }
  },
  "model": "zai/glm-5.2",
  "small_model": "zai/glm-4-flash"
}

Why Critical¶

• Best price/performance for coding in RU/CN markets
• OpenAI-compatible — works with @ai-sdk/openai-compatible
• GLM-5.2 rivals Claude 4 Sonnet in coding benchmarks
• Free tier available

2. Moonshot — Kimi Series¶

Current Models (July 2026)¶

API Endpoint¶

Base URL: https://api.moonshot.ai/v1
Auth: Bearer token (MOONSHOT_API_KEY)
Compatible: OpenAI API format

Integration¶

{
  "provider": {
    "moonshot": {
      "options": {
        "baseURL": "https://api.moonshot.ai/v1"
      }
    }
  }
}

Notes¶

• Project already has moonshot provider, but model name "K2.6" is incorrect
• Should be "Kimi K2" or "moonshot/kimi-k2"

3. Anthropic — Claude Series¶

Current Models (July 2026)¶

Integration¶

opencode has native @ai-sdk/anthropic SDK. No baseURL needed.

{
  "provider": {
    "anthropic": {}
  },
  "model": "anthropic/claude-4-opus",
  "small_model": "anthropic/claude-4-haiku"
}

Notes¶

• Project says "Claude 4" — should specify "Claude 4 Opus" or "Claude 4 Sonnet"
• Anthropic has special headers for interleaved thinking + fine-grained tool streaming

4. OpenRouter — Aggregator¶

Why Important¶

• Single API key → access to 100+ models
• Fallback chains across providers
• Price optimization
• Works with @openrouter/ai-sdk-provider (native in opencode)

Integration¶

{
  "provider": {
    "openrouter": {}
  },
  "model": "openrouter/anthropic/claude-4-opus",
  "small_model": "openrouter/deepseek/deepseek-v4-flash"
}

5. Alibaba — Qwen Series¶

Current Models (July 2026)¶

Integration¶

opencode has native @ai-sdk/alibaba SDK.

{
  "provider": {
    "alibaba": {}
  }
}

6. opencode Plugin Ecosystem¶

Plugin Architecture (from opencode docs)¶

Plugins are JS/TS modules that hook into events: - Tool events: tool.execute.before, tool.execute.after - Session events: session.created, session.idle, session.compacted - File events: file.edited, file.watcher.updated - Message events: message.updated, message.part.updated - TUI events: tui.prompt.append, tui.command.execute, tui.toast.show - Shell events: shell.env - Permission events: permission.asked, permission.replied - LSP events: lsp.client.diagnostics, lsp.updated - Compaction hooks: experimental.session.compacting

Plugin Loading¶

1. Global config (~/.config/opencode/opencode.json) — npm packages
2. Project config (opencode.json) — npm packages
3. Global plugin dir (~/.config/opencode/plugins/) — local files
4. Project plugin dir (.opencode/plugins/) — local files

Custom Tools¶

Plugins can add custom tools via tool() helper with Zod schema.

Best Practices¶

1. Use client.app.log() instead of console.log() for structured logging
2. Use Bun shell API ($) for command execution
3. TypeScript plugins get type safety via @opencode-ai/plugin
4. Local plugins need package.json in config dir for npm dependencies
5. Plugins can inject env vars into all shell executions via shell.env hook

7. Recommended Provider Updates¶

Add to registry¶

[zai]="ZAI_API_KEY|--zai-key|z.ai GLM-5.2|yes"
[openrouter]="OPENROUTER_API_KEY|--openrouter-key|OpenRouter (100+ models)|yes"
[alibaba]="ALIBABA_API_KEY|--alibaba-key|Alibaba Qwen3|yes"
[deepinfra]="DEEPINFRA_API_KEY|--deepinfra-key|DeepInfra (fast inference)|yes"

Update model names¶

[xai]="XAI_API_KEY|--xai-key|xAI Grok 4|no"           # Grok 3 → Grok 4
[moonshot]="MOONSHOT_API_KEY|--moonshot-key|Moonshot Kimi K2|no"  # K2.6 → K2
[anthropic]="ANTHROPIC_API_KEY|--anthropic-key|Anthropic Claude 4 Opus|no"  # Claude 4 → Claude 4 Opus

Default model recommendation¶

{
  "model": "zai/glm-5.2",
  "small_model": "zai/glm-4-flash",
  "provider": {
    "zai": {
      "options": {
        "baseURL": "https://api.z.ai/api/paas/v4"
      },
      "fallback": ["deepseek", "opencode"]
    }
  }
}

8. Local LLM Backends (Isolated Circuit)¶

Already well-covered in project: - Ollama (:11434) — primary - LiteLLM (:4000) — proxy/gateway - vLLM (:8000) — high-throughput - SGLang (:30000) — structured generation

Recommended local models¶