grok-chat-model-map

# Grok Chat Model Map

Canonical **Chat mode** skill for [grok.com](https://grok.com) Chat (also iOS / Android / X Grok).

**Verified picker (2026-09-06 ET):** `Auto` · `Fast` · `Expert` · `Heavy` · `Build`

Begin every activation with: **"Grok Chat model map locked…"**  
Then emit the pick card (Mode / Why / Not).

Supporting files:
- `references/CHEAT_SHEET.md` — paste-ready user card
- `references/mode-catalog.md` — modes + API ids + aliases
- `references/PASTE_PACK.md` — short share paste blocks
- `README.md` — install into `~/.grok/skills`

## Hard rules

1. **Never invent picker labels.** Only: Auto, Fast, Expert, Heavy, Build.
2. **Never send Chat mode names as API `model` strings.** `Auto` / `Expert` ≠ `grok-4.6`.
3. **Grok 4.6** (`grok-4.6`) is the Chat/Code generation on [docs.x.ai/models](https://docs.x.ai/developers/models) (“Chat: Grok 4.6” / “Code: Grok 4.6”) — **not** a fifth depth tile in the verified Chat picker. Docs also: Images → Image **2.0**, Videos → Video **1.5**.
4. **Build** in the Chat picker is a **surface hop** to Grok Build, not a reasoning depth.
5. **Imagine** stills/video → skill `imagine-model-overrides` (sidebar Imagine). Docs default stills: **`grok-imagine-image-2.0`** (Quality Mode); not this skill.
6. **Cinematic Studio** `tools/models.py` is a different registry (do not merge).

## Mode catalog (Chat UI)

| Mode | Kind | When to use | Relation to Grok 4.6 |
|------|------|-------------|----------------------|
| **Auto** | Router | Everyday default | Routes Fast-like vs Expert-like; not an API id |
| **Fast** | Latency path | Quick facts, short asks | Same Chat product family; lighter route — API id not shown in UI |
| **Expert** | Deep reasoning | Hard analysis, long reasoning | Best match for flagship Chat on **4.6** generation |
| **Heavy** | Multi-agent max | Hardest multi-step depth | Separate Heavy stack (not “slow Expert”) |
| **Build** | Product switch | Coding agent / repo work | Opens Grok Build; Build default intelligence also **grok-4.6** |

Full tables: `references/mode-catalog.md`.

## Argument shortcuts

| Arg | Action |
|-----|--------|
| `auto` / `fast` / `expert` / `heavy` / `build` | Lock that mode + why |
| `tree` | Print surface decision tree only |
| `api` | Print API id reminder (never mode names as `model`) |
| _(empty)_ | Ask goal in one line, then recommend |

## Quick picks

| You want… | Pick |
|-----------|------|
| Default / mixed day | **Auto** |
| Speed | **Fast** |
| Depth / hard reasoning | **Expert** |
| Max multi-agent depth | **Heavy** |
| Coding agent | **Build** |
| Stills / video | **Imagine** (other skill) |
| HTTP API call | API id e.g. `grok-4.6` — not a Chat mode |

## Surface tree

```
Chat thread          → Auto | Fast | Expert | Heavy
Chat picker → Build  → Grok Build coding agent
Sidebar Imagine      → Imagine image/video models
api.x.ai / SDK       → grok-4.6, grok-4.5, …
Grok Build TUI       → /model aliases (host config)
Studio registry      → tools/models.py (separate)
```

## Output card (always)

```markdown
## Grok Chat pick
- Mode: …
- Why: …
- Not: API id / Imagine / Studio registry
```

## Tier note

Verified session 2026-09-06: all five modes visible with **no lock badges**. Quotas can still differ by plan (Heavy often needs SuperGrok Heavy for full compute).

## References

- Plan / pack PR: https://github.com/FineComputer14451/Grok-Imagine-Cinematic-Studio/pull/48
- Docs: https://docs.x.ai/developers/models · https://docs.x.ai/developers/grok-4-6
- Sibling skill: `imagine-model-overrides` (Imagine Quality/Fast · Video 1.0/1.5)