Elon just gave us a pretty clear Grok roadmap.
4.7: roughly Opus 5.0, with multimodal still needing work. 4.8: noticeable improvement. 4.9: probably Astra/Fable class. Grok 5: maybe better than anything. 4.7 looks like the competitive step. 4.8 through 5 are where Elon
If anyone wants to compare the big models this year, the situation has become interesting once again 🧠
Your options are: Gemini 3.8 Flash, GPT 5.6 Sol, Opus 5, and Kimi K3. Each one of them is strong in its own way, but the difference depends on exactly what you want from it.
You can make games with Unity, right? So I made a potato digging game for Claude's Opus 5 and gave it a try...
result. Are you walking? →Do you have potatoes? →Sell? There's something strange about the movement lol Even so, Blender's MCP didn't connect properly until the end 🤣 What we ended up with is a mysterious game 😂 NaniColle lol
OpenAI and Anthropic are now separated by 0.53 points on the leaderboard.
Top 5 AI models by real benchmark score - September 2026 Claude Fable 5.1 - 84.61, $10/$50 per M tokens GPT-6 Astra - 84.08, $10/$50 per M tokens Claude Opus 5 - 81.97, $5/$25 per M tokens Claude
You have reached the end of the archive
All of Claude Opus 5