Model Router Update Models

Pull latest community model recommendations from GitHub.

Trigger

User runs /model-router:update-models or /model-router:update-models --reset

Arguments

• --reset (optional): Reset to community defaults, discarding local customizations

Instructions

1. Fetch Latest Community Recommendations

# Fetch latest models.yaml from GitHub
curl -s https://raw.githubusercontent.com/mikemorgenstern/model-router/main/data/models.yaml > /tmp/models-latest.yaml

If fetch fails:

Could not fetch latest model recommendations.
Check your internet connection or try again later.

Current models file: ~/.model-router/models.yaml

2. Compare Versions

Read current ~/.model-router/models.yaml and compare with latest:

import yaml

# Load both files
current = yaml.safe_load(open("~/.model-router/models.yaml"))
latest = yaml.safe_load(open("/tmp/models-latest.yaml"))

# Compare versions
current_version = current.get("version")
latest_version = latest.get("version")

# Check for differences in:
# - Task type recommendations (best/runner_up/budget)
# - Model definitions (new models, updated benchmarks)
# - Cost tiers

3. Show Changes

Community Model Recommendations Update
═══════════════════════════════════════════════════════════

Your version: 1.0.0
Latest version: 1.2.0

Changes since your version:
───────────────────────────────────────────────────────────

NEW MODELS:
  • gpt-5.3 (ChatGPT Plus)
    HumanEval: 94.5%, better than GPT-5.2

  • gemini-3.5-flash (Google AI Pro)
    Replaces gemini-3-flash-preview

UPDATED RECOMMENDATIONS:
  • code_review: deepseek-r1-70b added as budget option
    Reason: Recent benchmarks show strong coding performance

  • reasoning: claude-opus-4.5 benchmarks updated
    GPQA: 91.2% → 93.5%

DEPRECATED:
  • gemini-2.0-flash → Use gemini-3-flash-preview instead
    (Will continue to work but not recommended)

COST CHANGES:
  • claude-haiku-4.5: $0.25/$1.25 → $0.20/$1.00 per 1M tokens
    (Price decrease)

═══════════════════════════════════════════════════════════

4. Handle Local Customizations

If user has made local customizations (via /model-router:learn or manual edits):

⚠ You have local customizations:
  • code_review: best changed to gpt-5.2 (community default: claude-sonnet-4.5)
  • reasoning: budget changed to llama-3.3-70b (community: gemini-flash)

Update strategy:
○ Merge updates (preserve your customizations, add new models)
○ Reset to community defaults (discard customizations)
○ Review changes (see details before deciding)

If "Merge updates":

• Add new models to the models section
• Update benchmarks for existing models
• Preserve user's task type preferences
• Add new task types from community

If "Reset to community defaults":

• Replace entire file with latest version
• Warn about losing customizations

If "Review changes": Show side-by-side comparison and let user choose per-item.

5. Apply Update

If user approves, update ~/.model-router/models.yaml:

# Backup current file
cp ~/.model-router/models.yaml ~/.model-router/models.yaml.backup

# Apply update (merge or replace)
cp /tmp/models-latest.yaml ~/.model-router/models.yaml

# If merge, apply customizations on top

Add update metadata:

# Updated from community recommendations
# Last updated: 2026-02-03
# Version: 1.2.0
# Your customizations: [preserved/reset]

6. Confirm Update

Model recommendations updated!

Version: 1.0.0 → 1.2.0

Applied changes:
  ✓ Added 2 new models (gpt-5.3, gemini-3.5-flash)
  ✓ Updated 3 task recommendations
  ✓ Updated 5 model benchmarks
  ✓ Preserved your local customizations

Backup saved: ~/.model-router/models.yaml.backup

Run /model-router:recommend to see updated suggestions.

7. Reset Option (if --reset flag)

This will reset your model recommendations to community defaults.

You will lose these customizations:
  • code_review: best=gpt-5.2 (custom) → claude-sonnet-4.5 (default)
  • reasoning: budget=llama-3.3-70b (custom) → gemini-flash (default)

Your tracking data will be preserved.
Backup will be saved to: ~/.model-router/models.yaml.backup

Continue?
○ Yes, reset to defaults
○ No, cancel

If yes:

cp ~/.model-router/models.yaml ~/.model-router/models.yaml.backup
cp /tmp/models-latest.yaml ~/.model-router/models.yaml

8. Show What's New

After update, highlight notable changes:

What's New in Model Recommendations 1.2.0:
──────────────────────────────────────────────────────────

🆕 GPT-5.3 now available for ChatGPT Plus subscribers
   HumanEval: 94.5% (improvement over GPT-5.2)
   Try: /model-router:test gpt-5.3

📈 DeepSeek R1 70B now recommended for coding tasks
   Strong performance on recent benchmarks
   Free local option for code review

💰 Claude Haiku price drop: 20% cheaper
   Now $0.20/$1.00 per 1M tokens

See full changelog:
https://github.com/mikemorgenstern/model-router/releases/v1.2.0

9. Suggest Testing

Want to test the new models?

○ Test gpt-5.3 now (/model-router:test gpt-5.3)
○ Test all new models
○ Skip for now