Benchmark-style model comparison graphic for different AI tasks

I don't pick GPT-6 by its score alone. Here's where I'd start with these 8: 1. Opus 5.5: proposals, reports and coding. 2. Fable 5.1: stress-test plans and drafts. 3. GPT-6 Astra: campaign strategy and research. 4. GPT-6 Sol: landing pages and workflows. 5. Grok 4.7: research with live search tools. 6. Kimi K3: large file sets and reports. 7. Gemini 3.8 Flash: analyse images and video. 8. GPT-6 Luna: tag feedback and extract data. But the job matters as much as the ranking (especially when you need a quick answer). The large numbers are Artificial Analysis benchmark scores. The smaller /10 ratings are editorial judgements (not my own test results). And those benchmarks use different effort settings, so treat this as a starting point for your next task. Try the same brief in two models, then compare what you can use without rewriting it (keep the brief identical). Repost ♻️ to help someone in your network.

— Charlie Hills (@charliejhills) Sep 23, 2026