# OpenCode Go Model Discovery (May 2026) ## Architecture Overview There are **two separate API surfaces** for OpenCode Go models, and they behave differently: | Access path | Endpoint | Auth source | Build agent support | Model naming | |---|---|---|---|---| | OpenCode CLI (built-in) | Internal (not user-configurable) | `auth.json` → `providers.opencode.apiKey` | Only `big-pickle` | `opencode/big-pickle` | | Direct OpenAI-compatible API | `https://opencode.ai/zen/go/v1` | Same API key, Bearer token | N/A (not CLI) | `qwen3.6-plus` (no prefix) | The built-in `opencode` provider and the direct Zen/Go API both use `@ai-sdk/openai-compatible` internally. The Go API at `https://opencode.ai/zen/go/v1` CAN be used directly as an OpenCode custom provider — the key requirement is **dual-auth provisioning** (see the "Adding Go Models via Custom Provider" section below). A 401 error usually means the API key is missing from one of the two required locations, not a server-side block. ## Go Subscription ($10/mo) — CLI `opencode models opencode` or `opencode models | grep "^opencode/"` shows: ``` opencode/big-pickle ← the ONLY Go subscription model the build agent accepts opencode/deepseek-v4-flash-free ← free tier (no subscription needed) opencode/minimax-m2.5-free ← free tier opencode/nemotron-3-super-free ← free tier opencode/qwen3.6-plus-free ← free tier ``` The build agent is server-side locked to only allow `big-pickle` for Go subscribers. The `-free` models are the free tier (no API key needed). Adding models to `provider.opencode.models` in `opencode.json` makes them appear in `opencode models` but the build agent rejects them with "Model not supported." ## Go Subscription — Direct API (via Hermes opencode-go provider) The raw Go API at `https://opencode.ai/zen/go/v1/models` returns **15 models** that all work with Bearer token auth: ``` minimax-m2.7 ← confirmed working minimax-m2.5 kimi-k2.6 kimi-k2.5 glm-5.1 glm-5 deepseek-v4-pro deepseek-v4-flash qwen3.6-plus qwen3.5-plus mimo-v2-pro mimo-v2-omni mimo-v2.5-pro mimo-v2.5 hy3-preview ← not in Hermes catalog (added manually) ``` Smoke-test with: ```bash curl -s --max-time 10 -H "Authorization: Bearer $KEY" \ "https://opencode.ai/zen/go/v1/models" | python3 -c \ "import sys,json; [print(m['id']) for m in json.load(sys.stdin)['data']]" ``` ## Adding Go Models via Custom Provider (OpenCode CLI) The Go API at `https://opencode.ai/zen/go/v1` can be added as a custom provider in `opencode.json` so all 13+ Go subscription models are available via `opencode run --model opencode-go/`. **Critical: dual-auth provisioning required.** The API key must be in **two places**: ### 1. Auth.json entry Add the provider name to `~/.local/share/opencode/auth.json`: ```json { "providers": { "opencode": { "apiKey": "sk-..." }, "opencode-go": { "apiKey": "sk-..." } } } ``` ### 2. Config options In `~/.config/opencode/opencode.json`, the same key must appear in `options.apiKey`: ```json { "provider": { "opencode-go": { "npm": "@ai-sdk/openai-compatible", "name": "OpenCode Go", "options": { "baseURL": "https://opencode.ai/zen/go/v1", "apiKey": "sk-..." }, "models": { "qwen3.6-plus": { "name": "Qwen 3.6 Plus", "limit": { "context": 128000, "output": 65536 } }, "minimax-m2.7": { "name": "MiniMax M2.7", "limit": { "context": 128000, "output": 65536 } }, "kimi-k2.6": { "name": "Kimi K2.6", "limit": { "context": 128000, "output": 65536 } }, ... } } } } ``` **Without BOTH entries:** Omitting the `auth.json` entry → `Missing API key`. Omitting `options.apiKey` → `Invalid API key`. Both must be present. ### Verification ```bash opencode models | grep "^opencode-go/" opencode run 'say OK' --model opencode-go/qwen3.6-plus ``` ### Models discovered by OpenCode CLI (13 of 15) When configured correctly, these models appear in `opencode models`: `qwen3.6-plus`, `qwen3.5-plus`, `minimax-m2.7`, `minimax-m2.5`, `kimi-k2.6`, `kimi-k2.5`, `glm-5.1`, `glm-5`, `deepseek-v4-pro`, `deepseek-v4-flash`, `mimo-v2.5-pro`, `mimo-v2.5`, `hy3-preview` Two API models (`mimo-v2-pro`, `mimo-v2-omni`) are listed in the API but don't appear in `opencode models` due to OpenCode's internal model discovery filtering. ## Model Name Mapping: CLI vs Direct API | CLI name | Direct API name | Status | |---|---|---| | `opencode/big-pickle` | ❌ not on API | Use via `opencode run` | | `opencode/qwen3.6-plus-free` | `qwen3.6-plus` | ✅ works via Hermes | | `opencode/minimax-m2.5-free` | `minimax-m2.7` | ✅ works via Hermes | | `opencode/deepseek-v4-flash-free` | ❌ not on API | Free tier, not Go sub | | `opencode/nemotron-3-super-free` | ❌ not on API | Free tier, not Go sub | | (not in CLI) | `kimi-k2.6`, `glm-5.1`, etc. | ✅ Go sub, Hermes only | **Key rule:** Model names with the `opencode/` prefix are CLI-only. Short bare names are API-only. Never use the `opencode/` prefix when hitting the direct API. ## Hermes Integration The `opencode-go` provider in Hermes taps the same Go subscription via `https://opencode.ai/zen/go/v1`. To configure: ```bash # 1. Copy key from OpenCode's auth store export KEY=$(python3 -c "import json; print(json.load(open('$HOME/.local/share/opencode/auth.json'))['providers']['opencode']['apiKey'])") # 2. Add to Hermes .env sed -i '/^OPENROUTER_API_KEY=/a OPENCODE_GO_API_KEY='"$KEY" ~/.hermes/.env # 3. Register provider in config.yaml # providers: # opencode-go: # api_key_env: OPENCODE_GO_API_KEY # base_url: https://opencode.ai/zen/go/v1 # 4. Use in-session hermes chat -q "test" --provider opencode-go -m qwen3.6-plus ``` ## Zen Free Models (No Subscription Required) The Zen API at `https://opencode.ai/zen/v1` also hosts **3 free models** that don't require any Go subscription — just the same OpenCode API key: - `deepseek-v4-flash-free` ✅ - `minimax-m2.5-free` ✅ - `nemotron-3-super-free` ✅ Other models listed at the Zen endpoint (`ring-2.6-1t-free`, `trinity-large-preview-free`, `qwen3.6-plus-free`) are either defunct or require Go credits. Only the 3 above work as true free models as of May 2026. ### Adding Zen Free Models as a Custom Provider (OpenCode CLI) Same dual-auth pattern as the Go provider: **auth.json:** ```json { "providers": { "opencode-zen": { "apiKey": "sk-..." } } } ``` **opencode.json:** ```json { "provider": { "opencode-zen": { "npm": "@ai-sdk/openai-compatible", "name": "OpenCode Zen (Free)", "options": { "baseURL": "https://opencode.ai/zen/v1", "apiKey": "sk-..." }, "models": { "deepseek-v4-flash-free": { "name": "DeepSeek V4 Flash (Free)", "limit": { "context": 128000, "output": 65536 } }, "minimax-m2.5-free": { "name": "MiniMax M2.5 (Free)", "limit": { "context": 128000, "output": 65536 } }, "nemotron-3-super-free": { "name": "Nemotron 3 Super (Free)", "limit": { "context": 128000, "output": 65536 } } } } } } ``` **Use:** `opencode run "..." --model opencode-zen/deepseek-v4-flash-free` ### Adding Zen Free Models as a Hermes Custom Provider **In config.yaml, under `custom_providers:`:** ```yaml - base_url: https://opencode.ai/zen/v1 api_key_env: OPENCODE_ZEN_API_KEY models: deepseek-v4-flash-free: context_length: 128000 minimax-m2.5-free: context_length: 128000 nemotron-3-super-free: context_length: 128000 name: zen-free ``` **In .env:** ``` OPENCODE_ZEN_API_KEY=sk-... ``` **Use in Hermes:** `/model custom:zen-free/deepseek-v4-flash-free` **Important:** Do NOT register `opencode-zen` as a standard provider in the `providers:` section — that pulls in Hermes' built-in catalog which lists paid Go/Zen models the user doesn't want in the picker. Use `custom_providers` with an explicit model list to keep only the free ones. ## Recommendation | Use case | Path | Model | |---|---|---| | Multi-file refactors, PRs | `opencode run -m opencode/big-pickle` | `big-pickle` (CLI-only) | | Chat, quick tasks, tool use | `/model opencode-go/qwen3.6-plus` | Any Go API model | | Premium cloud models | `openrouter/~anthropic/claude-sonnet-latest` | OpenRouter