Use
GPT-6 Astra
OpenAI
The top creative writing rating on EQ-Bench (2173), at about $95.23 per 1,000 pieces.
Benchmarks tell you who's ahead. Dervity tells you what to use: the best model for your work, and what you give up if you go cheaper or faster.
Each dot is an AI model. Higher is smarter, further right costs more per task. Anything under the green line has a cheaper model that's at least as smart.
The best deal at each price
Intelligence: Artificial Analysis Intelligence Index at each model's strongest setting that answers within 45 seconds. Cost per task includes thinking tokens and is on a log scale.
Pick the job. Each guide ranks every model for it and names the one to use, plus a cheaper option that's nearly as good.
Where the newest models land, what they replace, and which jobs they're now the pick for.
Claude Opus 5.5
Anthropic · September 2026 · replaces Claude Opus 5
Claude Fable 5.1
Anthropic · September 2026 · replaces Claude Fable 5
GPT-6 Astra
OpenAI · September 2026
Muse Spark 1.3
Meta · September 2026
Grok 4.7
SpaceXAI · September 2026 · replaces Grok 4.6
MiMo-V2.6-Pro
Xiaomi · September 2026