Claude Haiku 5.5 vs Nemotron 3.5 Super VL
Comparer des modèles
Claude Haiku 5.5 is more worth it — 64% vs 42%
How we score worth-it? →Anthropic's cheapest model, with an effort-control dial for reasoning.
- Previously we used Gemini flash 3.8 as it was capable and affordable. However, haiku 5.5 just blows it out of the water. It's more than 10x cheaper, 2x faster and finds 3x as many real issues.— hn ↗
- In our evals it passes more cases than Sonnet 4.6, at about a twenty-fifth of the cost and half the latency.— github ↗
- This dirt cheap Haiku is better than a frontier from 3 months ago 😮— youtube ↗
- Haiku is a bit more expensive than Luna on < 100k token, but I just don't have any agentic work below 100k, so this is going to be 5x more expensive than shown.— hn
“Noticeably smarter than GPT-6 Luna — beats Sonnet 5 in some coding benchmarks”
“Intelligence per dollar has grown dramatically in a few months”
“100k-token cutoff is absurdly low for agentic work”
Open 120B vision-language model — 5x faster for always-on agents
- Post-training an open-weight model is now an afternoon's work. Accuracy on that task rose from 17% to 78% after 3.3 hours of training on four NVIDIA H200 GPUs.— blog ↗
- It's a very interesting size range to explore the space of capabilities. Appreciate having something with more than 6B active params in this size range.— huggingface ↗
- When the pedants want to talk open *source* vs open weight, Nemotron models are the only ones to fit the real definition and not just be some experiment uploaded to HF by some rando. They're actually useful.— reddit ↗
- Super VL (120B, 12B active, NVFP4): the full 1,048,576-token window on one box. A code hidden at 10, 50 and 90 % of prompts up to a million tokens was found 12 times out of 12. About 70 tok/s rewriting code with one request, 82 tok/s in all at 4 requests.— nvidia-dev-forum ↗
- Their own benchmarks seem to show it well behind Qwen 3.8 flash on pretty much everything except generation speed.— hn
- No suitable use case yet. Lots of tool call errors. It does not strictly follow instructions and contracts, which I need for unattended jobs.— other ↗
- So it's actually 120B — I'd have taken it for a 16B model. Other 100B+ models are way ahead; why is Qwen 3.8 Flash Next so much better? I'm genuinely puzzled.— bilibili ↗
- Every nemotron model ive ever used. Flash, ultra, lightning. Every single model is significantly worse and impossible to use vs another model in the same weight range.— reddit ↗
- Over 666 streamed requests with max_tokens about 261,700 (262,144 minus the prompt), 123 ended with finish_reason: "length" at exactly 131,072 completion tokens— nvidia-dev-forum ↗
Which should you pick?
The gap is clear: Claude Haiku 5.5 sits at 64% while Nemotron 3.5 Super VL is at 42%, a 22-point difference. If you just want the safer pick, start with Claude Haiku 5.5.
On entry price, Claude Haiku 5.5 is cheaper ($0.1 vs $0.15, Nemotron 3.5 Super VL costs about 50% more).
Both are in the Chat category with similar coverage; the real difference is stability and cost — run one of your own tasks through each before deciding.
What users actually said about Claude Haiku 5.5: "Previously we used Gemini flash 3.8 as it was capable and affordable. However, haiku 5.5 j…"
FAQ
Is Claude Haiku 5.5 or Nemotron 3.5 Super VL more worth it?
Claude Haiku 5.5 is more worth it, with a 64% worth-it score vs 42% for Nemotron 3.5 Super VL.
How much does Claude Haiku 5.5 cost vs Nemotron 3.5 Super VL?
Claude Haiku 5.5: $0.10/$0.50 per 1M (≤100k) · Nemotron 3.5 Super VL: $0.15/$0.50 per 1M.
What do users say about Claude Haiku 5.5 vs Nemotron 3.5 Super VL?
Claude Haiku 5.5: Noticeably smarter than GPT-6 Luna — beats Sonnet 5 in some coding benchmarks. Nemotron 3.5 Super VL: no highlights yet.