🥇 '@Bitdeer_AI is one of the lowest-cost providers for DeepSeek V4 Flash, sitting in the most attractive quadrant while remaining competitive on speed according to the latest @ArtificialAnlys benchmark. 🦾Our goal with this pricing is to accelerate the adoption of cutting-edge intelligence. We hold the advantage through full vertical integration: from owned GPUs down to token serving, backed by deep optimization capabilities. ،🚀Sign up, get your $5 voucher and use Model Studio now: @deepseek_ai ⚡ #OpenModels#

Open models are running hot on @Bitdeer_AI Cloud! 🔥 Join us tomorrow evening at @Lorong_AI to learn how builders are creating production-ready AI agents. We’re moving beyond the hype with live demos of @Bitdeer_AI Model Studio powering: ⚲ ERP system automation ⚲ WhatsApp, Telegram & Email interactions ⚲ Human-in-the-loop control for automated ERP workflows. ⚲ Native, no-code integration with @Bitdeer_AI Model Studio. ✈️ Stop experimenting. Start deploying. Grab your spot and let's put these models to work! 👇

Open models have more than doubled their share of Vercel AI Gateway tokens: 29% in June to 61.3% in this September-end snapshot. Closed models fell from 71% to 38.7%. Meanwhile, Ornn’s September 30 data shows H200 rental rates up 26% over 30 days and B200 up 33%. H100 and A100 fell 9% and 14%. From late July through September, tracked blended token prices fell 61% for OpenAI, 63% for DeepSeek, 22% for Anthropic and 15% for Google. Interesting...

Open models are no longer losing at the top of the funnel on benchmark-performance; they are losing at the last mile. 79% of developers use them, but only 51% reach production vs 63% for closed models, vendor-backed deployments reach production at 67% versus 33% for internal builds, and the closed-open production gap actually widens from 1 point in small firms to 16 points in enterprises. People are dropping them because they're harder to run reliably.

Open models are ~50x cheaper than closed ones. Mozilla CTO @raffi talks about the economics of open source AI, the "amnesiac" problem with agent memory, and why concentration risk isn't just about who's closed — it's about who's open, too. Read/watch here:

Open-weight AI labs have caught up to frontier models, and their low prices are eating up serious market share. The only real narrative holding them back right now is the fear of China stealing user data. If you are worried about that, you have two simple choices: Run the open models locally, or use a US-based third-party inference provider.

Open models do not just create more builders. They create more agent attempts. More experiments. More tool calls. More failed runs. More inference demand. More need to verify what actually happened. The open-source wave expands the execution market, not just the model market.

Open source is closing the gap with closed labs, and fast. At ICML, @tri_dao broke it down: open models are now just months behind on quality, and the serving stack (kernels, inference engines, speculative decoding) keeps compounding. Full clip 👇🏻

Open models recap: more on Kimi K3, Qwen 3.8, Xi's WAIC speech, distillation, the open-closed gap, and what's next A podcast with Florian Brand. YouTube: Interconnects:

Most open models were trained to repeat one government's talking points. Nous Research just showed it takes one fine-tune to fix that. Arete-27B is Qwen3.8-27B retrained to answer straight on free speech, democracy and privacy: - Propaganda rate went from 22% to 1% on an 803-prompt test - Only about 53M tokens of fine-tuning - No drop in general ability on their benchmark - Weights, eval and training data are all open "The values in an open model are one fine-tune away."