NVIDIA's NEW Nemotron 3.5 Bolt 4B Model Is INSANE... I Tested It 10 Ways On An 8GB Laptop

NVIDIA's new Nemotron 3.5 Bolt E4B (8B total, 4B effective) installed on my own Windows laptop with an RTX 4060 8 GB, and put through 10 real tests: rule following, tool calls, agent chains, JSON, long context, vision, thinking mode, NPC roleplay and code. Real screen recordings, real logs, and the failures are shown. Result: 5 clean passes, 4 partial, 1 fail. Measured here: about 77 tokens/s and about 0.2 s to first word (warm), 64K context loads on 8 GB at 7.8 of 8.2 GB. NVIDIA's own benchmark table (their harness) also has Qwen3.5 4B ahead on tool use and 64K long context. I did not test Qwen. Run it: ollama run nemotron-3.5-bolt:e4b (needs Ollama 0.40.10 or newer) Model card: https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB Ollama page: https://ollama.com/library/nemotron-3.5-bolt Chapters 0:00 Is NVIDIA's new Bolt 4B any good? 0:39 Intro 0:48 What Nemotron 3.5 Bolt claims 1:47 Install on Windows with Ollama 2:30 How the 10 tests work 2:47 Test 1: rules across turns 3:38 Test 2: calculator tool calls 4:10 Test 3: agent chain (and a trap) 4:43 Test 4: knowing when not to use a tool 5:07 Test 5: messy text to JSON 5:47 Test 6: 64K context on 8 GB 6:55 Test 7: reading a chart 7:43 Test 8: thinking on vs off 8:36 Test 9: NPC under attack 8:59 Test 10: code vs hidden tests 9:42 Speed and memory 10:05 Against Spark X2.6 4B 11:05 Scorecard and verdict Cheap and reliable GPUs: RunPod https://get.runpod.io/pe48 Notes: tests use temperature 0.6, seed 7, 16K context, Ollama 0.40.11, Q4_K_M. Small samples, one machine. The presenter intro and outro are AI-generated (Seedance lip-sync from the creator's likeness and cloned voice). 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/