Skip to content

MODELS.md

Latest commit

 

History

History
185 lines (148 loc) · 8.2 KB

MODELS.md

File metadata and controls

185 lines (148 loc) · 8.2 KB
Raw

Models

Default and fallback model assignments for every harness in this repo.

Source of truth

models.json maps a stable alias to a concrete model ID. Templates (*.tpl.*) reference the alias as an __UPPER_SNAKE__ placeholder; scripts/llm-update.sh sed-substitutes them into the generated configs.

make llm-update

Never hand-edit a generated config. Edit the .tpl.* file, regenerate, and commit both.

Aliases

Placeholder Model ID
__CLAUDE_OPUS__ claude-opus-5
__CLAUDE_SONNET__ claude-sonnet-5
__CLAUDE_HAIKU__ claude-haiku-4-5-20251001
__GPT__ gpt-5.6-sol
__GPT_LUNA__ gpt-5.6-luna
__GPT_IMAGE__ gpt-image-2
__GEMINI_PRO__ gemini-3.1-pro-low
__GEMINI_FLASH__ gemini-3.7-flash-high
__DEEPSEEK_FLASH__ deepseek-v4-flash
__DEEPSEEK_PRO__ deepseek-v4-pro
__GLM__ glm-4.7
__GEMMA__ gemma-4-31b-it
__GEMMA_LOCAL__ gemma3:4b
__MINIMAX__ minimax-m3
__KIMI__ kimi-k3
__GROK__ grok-4.5
__QWEN__ qwen3.6-plus
__QWEN_LOCAL__ qwen3.5-0.8b-optiq

This table is the complete set of keys in models.json. Every key also gets two derived forms: __<KEY>_PRETTY__ for the display name ("Deepseek V4 Flash") and __<KEY>_NONDOT__ for the dot-stripped ID, used in OpenRouter @preset/ names.

Provider-specific slugs remain separate even when they represent the same model family. __GEMINI_PRO__ and __GEMINI_FLASH__ are IDs verified through CLIProxy, while Antigravity's native model ID stays directly in its provider-owned template.

Two provider-specific overrides are declared in scripts/llm-update.sh rather than in models.json, because the upstream ID differs from the canonical one:

Placeholder Value Used by
__GPT_IMAGE_OPENROUTER__ openai/gpt-5.4-image-2 OpenRouter, aliased back to gpt-image-2
__DEEPSEEK_FLASH_0731__ deepseek-v4-flash-0731 Aliyun, aliased back to deepseek-v4-flash

The shared fallback chain

Harnesses that support runtime fallback use one chain, in this order:

deepseek-v4-flash  (primary)
  -> gemma-4-31b-it
  -> glm-4.7
  -> free           (OpenRouter free router, last resort)

Rules:

  • deepseek-v4-flash is the only DeepSeek model on any automatic path. deepseek-v4-pro stays in the CLIProxy catalog and is addressable by explicit request, but it is never a fallback hop and never a default.
  • Every hop resolves through CLIProxy, so provider-level rotation (OpenCode -> Aliyun -> OpenRouter) already happens inside a single hop. Do not add a hop that repeats the primary model.
  • Keep the chain model-diverse. Each hop should be a different vendor family so a vendor-wide outage cannot exhaust the chain.

Per-harness assignments

Fallback-capable

Harness Default Fallback chain Config
OpenCode shunkakinoki/deepseek-v4-flash shared chain, shunkakinoki/ prefix opencode-fallback.tpl.jsonc
OpenClaw cliproxy/deepseek-v4-flash shared chain, cliproxy/ prefix openclaw.tpl.json
Hermes cliproxy/deepseek-v4-flash shared chain via fallback_providers config.tpl.yaml
OMP cliproxyapi/deepseek-v4-flash shared chain, cliproxyapi/ prefix config.tpl.yml

OpenCode fallback is driven by the opencode-runtime-fallback@0.2.3 plugin:

Setting Value
retry_on_errors 401, 404, 429, 500, 502, 503, 504
retryable_error_patterns unknown provider for model
max_fallback_attempts 5
cooldown_seconds 60
timeout_seconds 30

OpenClaw and Hermes have no equivalent error-pattern matcher, so they hard-fail on the unknown provider for model <prefixed-name> 400 that OpenCode absorbs. OMP uses native retry.fallbackChains instead of a plugin.

Hermes also runs a Mixture-of-Agents preset: reference models deepseek-v4-flash + minimax-m3, aggregator deepseek-v4-flash.

No fallback chain

Harness Role Model Config
OpenCode small_model shunkakinoki/glm-4.7 opencode.tpl.jsonc
OpenCode code-reviewer agent shunkakinoki/deepseek-v4-flash opencode.tpl.jsonc
OMP smol, commit, task cliproxyapi/glm-4.7 config.tpl.yml
OMP slow, vision, plan cliproxyapi/deepseek-v4-flash
Codex default gpt-5.6-sol config.tpl.toml
Codex subagents gpt-5.6-luna
Codex qwen-local profile qwen3.5-0.8b-optiq (LM Studio)
Antigravity default gemini-3.7-flash-high (native Antigravity provider) settings.tpl.json
Pi defaultModel glm-4.7 (provider cliproxyapi) settings.tpl.json
Factory (droid) session default deepseek-v4-flash settings.tpl.json
Factory (droid) custom local gemma3:4b (Ollama)
aichat default cliproxy:glm-4.7 config.tpl.yaml
llm default glm-4.7 default_model.tpl.txt
Handy transcript post-process @preset/glm-4.7 (OpenRouter) settings_store.tpl.json

OMP selects only cliproxyapi/*. Subagent overrides: code-explorer, comment-analyzer, and pr-test-analyzer use glm-4.7; the rest (code-architect, code-reviewer, code-simplifier, silent-failure-hunter, type-design-analyzer) use deepseek-v4-flash. The registry in models.tpl.yml lists the CLIProxy aliases and discovers the rest from local /v1/models. OMP fallback uses the shared chain (deepseek-v4-flash -> gemma-4-31b-it -> glm-4.7 -> free).

Fish shortcuts

The l suffix means local (LM Studio), h means headless.

Function Model
ocxe, ocxeh cliproxyapi/deepseek-v4-flash
ocxel, ocxelh lmstudio/qwen3.5-0.8b-optiq
coxe, coxeh gpt-5.6-sol
coxel, coxelh qwen3.5-0.8b-optiq (--oss --local-provider lmstudio)
pixe, pixeh cliproxyapi/glm-4.7
pixel, pixelh lmstudio/qwen3.5-0.8b-optiq

CLIProxy routing

Three provider prefixes appear above. All three are CLIProxy. Each harness names its own, so the prefix alone does not tell you remote vs local. cliproxyapi/ in particular means different endpoints in different harnesses.

Prefix Harness Endpoint
shunkakinoki/ OpenCode https://cliproxy.shunkakinoki.com/v1 (remote)
cliproxy/ OpenClaw, Hermes https://cliproxy.shunkakinoki.com/v1 (remote)
cliproxyapi/ OpenCode http://localhost:8317/v1 (local)
cliproxyapi/ OMP http://127.0.0.1:8317/v1 (local)
cliproxyapi/ Pi https://cliproxy.shunkakinoki.com/v1 (remote)

Every one of them resolves through config.tpl.yaml. Higher priority wins.

Provider Priority DeepSeek models served
opencode 300 deepseek-v4-pro, deepseek-v4-flash
aliyun 200 deepseek-v4-pro, deepseek-v4-flash-0731 aliased to deepseek-v4-flash
openrouter 100 @preset/deepseek-v4-pro, @preset/deepseek-v4-flash

So a single deepseek-v4-flash request tries OpenCode Zen, then Aliyun, then OpenRouter before the harness-level fallback chain sees a failure.

Changing a model

  1. To swap a model version everywhere: edit models.json, run make llm-update, commit the template and generated files together.
  2. To change a default or a fallback hop for one harness: edit that harness's .tpl.* file, run make llm-update.
  3. Update the matching assertions in spec/llm_update_spec.sh, spec/openclaw_hydrate_spec.sh, spec/hermes_hydrate_spec.sh, and spec/cliproxyapi_spec.sh, then run shellspec.