This Week in AI Models: Claude Still #1, but Open-Source is Catching Up Fast

I check the rankings every week. This one had some moves worth talking about.

Claude is still king, but the lead is shrinking

Claude Opus 4.6 held #1 again. Weekly usage grew 12%, down from 18% the week before.

It’s not that Claude got worse. It’s that the alternatives got better. Claude still dominates on complex reasoning and long documents — nobody touches it there. But the gap isn’t what it was three months ago.

Open-source models are actually competing now

Mistral’s latest jumped from #21 to #16 this week. The growth is coming from code-related usage — developers are voting with their API keys.

DeepSeek V4 crept up too. Qwen held steady. Three open-source players, all climbing. Not a fluke.

What’s driving it? Price, mostly. Mistral’s pricing is aggressive, and for a lot of coding tasks, you don’t need Claude-level reasoning. You need fast, cheap, good enough. That’s where open-source is winning.

Prices keep dropping

Average price across the top 50 models dropped another 8% this month. The SiliconFlow / Groq price war is still going, and honestly, I don’t think it stops anytime soon.

If you’re doing high-volume inference, this is your moment. Cheaper models mean you can experiment more, run more evaluations, try things you’d have hesitated on before.

What I’d do

• Tight budget? Check SiliconFlow’s mid-tier open-source models. Best bang for your buck right now.
• Need the best? Claude Opus 4.6 on OpenRouter. You’ll pay more, but for quality-critical work it’s still the safest bet.
• Want speed AND savings? Groq. Faster than OpenRouter, noticeably cheaper.

Next week I’m watching whether Mistral cracks the top 15.