NVIDIA Nemotron 3.5 Lightning
Version: 3.5 Lightning
By NVIDIA
Released: 2026-08-11
NVIDIA's open-weights hybrid Mamba-2/Mixture-of-Experts/Attention model, released 11 August 2026, built specifically for high-volume, low-latency AI agent workloads. 31.6B total parameters with only 3.6B active per token, ships alongside NeMo Switchyard, an open-source router for multi-model agent workflows.
Visit NVIDIA Nemotron 3.5 LightningAI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
NVIDIA Model Timeline
512k tokens context
AI Evaluation
A fast, cheap, open-weights model purpose-built for agentic workloads rather than raw benchmark-topping intelligence, with NeMo Switchyard as a genuinely useful companion router for multi-model pipelines.
Pros
- Open weights, free to self-host
- Sparse MoE design (3.6B active of 31.6B total) makes it fast and cheap to run
- Ships with NeMo Switchyard for routing steps of an agent workflow to the best-suited model
Cons
- Optimised for speed/cost over peak raw intelligence versus larger frontier models
- Newer release with a shorter independent-benchmark track record
- Best value requires adopting NeMo Switchyard's routing approach, not just the base model alone
Related Tools
Claude Opus 4.8
Claude Opus 4.8 is Anthropic's June 2026 flagship model, succeeding Opus 4.7. It posts a headline score of 81 on the hardest agentic coding and reasoning suites, holds long-horizon tool-use plans together across far more steps, and is notably more candid about its own uncertainty - refusing to fabricate rather than confidently pressing on. It is the default choice for serious agentic and software-engineering workloads.
Claude Fable 5
Claude Fable 5 is Anthropic's most intelligent generally available model and the first of its Mythos-class tier, positioned above Opus. It tops the Artificial Analysis Intelligence Index at 60, leads SWE-bench Pro at 80.3%, and dominates knowledge-work benchmarks on substance - at $2.75 per measured task, the highest in the field. It returned to sale on 1 July 2026 after a fortnight-long US export-control suspension.
Claude Sonnet 5
Claude Sonnet 5 is Anthropic's mid-size default, released 30 June 2026 with a 1M-token context window as standard. It scores 53 on the Intelligence Index - three points off Opus 4.8 - and actually edges the flagship on GDPval knowledge work, at roughly half the per-token cost, with introductory $2/$10 pricing until 31 August 2026.