Rank#1
NewsProgression
DeepSeek-V4-Pro-0813 pricing is ridiculous
• $0.435 / 1M input • $0.87 / 1M output What the hell have they created
Views119KLikes1.8KReposts81Replies49Saves244
Lumina weekly report
Ten posts selected from the week’s strongest AI news, credible reports, leaks and model progress—shown with their original X media and engagement snapshot.
• $0.435 / 1M input • $0.87 / 1M output What the hell have they created
Google AI Pro for $4.99/month for a year Great value, but they don't have a frontier model so you have to hope Gemini 4 makes it worth it or your just paying for Cloud Storage
How is a 27B model even putting up numbers like this • SWE bench Pro: 61.7 • QwenSWEBench: 79.0 • CoWorkBench: 70.7 • LiveCodeBench: 90.3 It’s genuinely Opus 4.6 level while being small enough to run on a desktop😭 27B models should not be this good, what the hell did Qwen do
Massive upgrade, the watermark was genuinely one of the most annoying things about making images with Gemini
Still the same 743B base model, but with much more post training These are the biggest jumps from GLM 5.2: • TerminalBench 3.0: 4.6 → 28.3 • DeepSWE: 46.2 → 66.9 • AutomationBench: 26.2 → 48.2 • GDPVal: 1508 → 1769, beating everyone here • CyberGym: 84.5, ahead of Fable + Sol What the hell did they just create
At this point I don't think they will either. Codex > Claude Code and it's not even close.
You could write something by yourself then make Claude proof read it or translate it, and the final text will still have Claude’s watermark • Copy/pasting doesn't remove it • Light editing won’t remove it • Translations will be watermarked • A detection API is coming too I can already see this causing so many problems, especially for students😂
Opus 5 feels trash and Fable 5 is nearly tied with Grok 4.6 which costs almost 4x less per task Grok 4.7 also comes out in a couple weeks which will decimate it Crazy week in AI
1.6T parameter / 49B active model is now on Hugging Face.
Looks like the wider release is basically imminent now, hopefully it drops tomorrow. From testing 'colossus' the codenamed model for Grok 4.6 in arena this looks like a very decent improvement overall.
