This 10 MILLION Token AI Is CRAZY... But Is It Real? (Mirendil Fern)

Mirendil just announced Fern 0.1, a model with a context window of up to 10 million tokens (about 23 MB of code, roughly half a million lines). I read all four Mirendil posts and their charts to separate what is shown from what is only claimed. Important: I did NOT run Fern. Access is "coming soon", so every number in the video is Mirendil's own, and the video says so. What's in the video: - The claim: 10M tokens, one researcher, 3.5 months, under $10M compute, over 99% of the code written by agents - Who Mirendil is: 20 founders from Anthropic, xAI, Google DeepMind and OpenAI; $200M seed; Google Cloud partnership - Fern's own figures: bits-per-byte keeps improving to 10.2M tokens (other models only measured to about 1M); RULER retrieval is 100% single-needle at 10M but multi-hop QA drops from about 65% to 45% - Mirendil-Base vs open models on 7 new repos (average bits per byte 0.222 vs 0.242 to 0.247; Kimi-K3 is better at 0.193) - Fern 0.1 on agentic tests: SWE-Bench Verified 73.4, Terminal Bench 6.1, well behind GLM-5.3, Kimi-K3 and Claude Opus 5.5 - What to watch next: access, full long-context results, outside reproduction Sources - https://mirendil.com/news/introducing-fern/ - https://mirendil.com/news/pretraining-at-scale-without-a-pretraining-team/ - https://mirendil.com/news/scaling-self-accelerating-ai-with-google/ - https://mirendil.com/news/announcing-mirendil/ RunPod: https://get.runpod.io/pe48 Chapters 0:00 10 million tokens: the claim 0:17 Why context size hurts today 0:57 Who is Mirendil 1:18 The Google Cloud compute deal 1:50 Fern 0.1 specs and what is missing 2:19 Fern's own charts: bits per byte and RULER 3:07 One researcher, 3.5 months, under $10M 4:02 The benchmark table 4:52 Fern on agentic coding tests 5:36 What to watch and my take Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/