Best for teams that want 30B-class reasoning without 30B-class hardware costs: engineers already running Nemotron-format pipelines get a drop-in upgrade, and cost-sensitive teams get open weights instead of a metered API. It leads 20B-class open rivals by double digits on PhD-level science reasoning and needs no metered API contract to run at scale.
Pulsar 16B is Multiverse Computing's open-weight reasoning model, compressed from NVIDIA's 30B-class Nemotron 3 Nano base into 3.1 billion active parameters using CompactifAI tensor-network technology. It keeps the source model's full context window and native tool-calling support, and ships as free, self-hostable open weights.
Where it sits
- --$ per 1M tokensBlended price (3:1)Lower is better-- / 64peer median $1.70/Mvendor price, checked by HokAI
- 4808 tok/stokens/sOutput speedHigher is better#1 / 39peer median 90 tok/scited: Artificial Analysis
- --% solvedSWE-bench VerifiedHigher is better-- / 28peer median 78.3%per source, see benchmark scores
- 71.4%% correctGPQA DiamondHigher is better#37 / 44peer median 88.3%per source, see benchmark scores
In the bottom third on GPQA Diamond (rank 37 of 44), and one of 65 whose vendor states it does not train on customer data.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: Multiverse Computing · Family: Pulsar
More about Multiverse Computing on HokAI
Context window: 1,000,000 tokens
Input modalities: text, tool-calls · Output: text, tool-calls
About Pulsar 16B
Pulsar 16B is an open reasoning model released on June 23, 2026 by Multiverse Computing, a Spanish AI infrastructure company headquartered in San Sebastian. Built in collaboration with NVIDIA, the model carries 16.15 billion total parameters with 3.1 billion active at any inference step. The architecture is a Hybrid Mamba2-Transformer with Mixture-of-Experts, compressed via Multiverse Computing's proprietary CompactifAI quantum-inspired tensor-network technology from the larger NVIDIA Nemotron 3 Nano 30B base model (31.6B total, 3.5B active). No retraining from scratch was involved; the compression preserved the original reasoning behavior, instruction-following, and tool-use interfaces intact.
On the AIME 2025 math reasoning benchmark, Pulsar 16B scores 87.22, within a fraction of a point of the 30B uncompressed Nemotron base and 15 points ahead of OpenAI's gpt-oss-20B. On GPQA Diamond, the PhD-level science reasoning benchmark, the model reaches 71.41, tracking the Nemotron base closely and outpacing gpt-oss-20B's 58.88 by over 12 points. Pulsar 16B leads gpt-oss-20B by 14 points on instruction-following (IFBench) and by 11 points on function-calling (BFCL-v4). Across standard reasoning, knowledge, coding, and tool-use benchmarks, Pulsar 16B matches its 30B-class starting point and outperforms gpt-oss-20B on nearly every axis.
Pulsar 16B inherits the 1,000,000-token context window from its NVIDIA Nemotron 3 Nano 30B base model. Multiverse Computing evaluated long-context recall using LongBench, AA-LCR, the RULER suite, and needle-in-a-haystack tasks at progressively longer spans. Needle retrieval remains essentially perfect on both sides of the 100K token mark. On harder RULER tasks at extended context, Pulsar 16B tracks the uncompressed base model closely, indicating the compression did not degrade long-context recall. The Nemotron base achieves RULER scores of 87.5% at 64K tokens, 82.92% at 128K, and 70.56% at 512K.
Pulsar 16B supports text input and output with full function-calling and tool-use interfaces retained from the Nemotron base. The model uses the same prompt format, reasoning interface, and tool-calling schema as the Nemotron 3 Nano family, making it a drop-in replacement in pipelines already using Nemotron-class models. Vision and audio capabilities are not present in the base Pulsar 16B release; those modalities exist in the broader Nemotron 3 Omni family but are not included in this checkpoint. Structured output generation and parallel tool calls are supported natively.
Because Pulsar 16B is released under the Apache 2.0 license, the weights are free to download and self-host. Inference costs reduce to hardware cost only. On an NVIDIA Blackwell GPU at 32 concurrent requests, the FP8 checkpoint delivers 4,808 tokens per second system throughput, a 43% improvement over the 30B base model's 3,363 tok/s. Time-to-first-token drops from 2.18 seconds to 1.24 seconds under the same hardware configuration. For teams preferring managed access, Multiverse Computing's CompactifAI API offers token-based pricing on AWS, Azure, and GCP, with costs reported up to 75% lower per token than equivalent frontier proprietary models for coding and reasoning tasks.
Pulsar 16B is available on Hugging Face under the MultiverseComputingCAI organization in BF16, FP8, and NVFP4 precision formats. BF16 requires approximately 32 GB VRAM, FP8 approximately 16 GB, and NVFP4 approximately 8 GB (excluding KV cache overhead at large contexts). The model was built and validated on NVIDIA Blackwell-class accelerated computing hardware. For teams not managing their own GPU infrastructure, the CompactifAI API via AWS Marketplace uses SageMaker HyperPod for serverless scaling. On-device deployment in NVFP4 is feasible on consumer hardware for short-context tasks.
Pulsar 16B inherits the safety alignment of the NVIDIA Nemotron 3 Nano base model. No separate system card specific to Pulsar 16B had been published as of June 2026. Multiverse Computing's technical documentation covers the compression methodology and evaluation setup but does not disclose specific refusal rates or red-teaming partnerships. The Apache 2.0 license permits modification of safety behavior, meaning users can adjust filtering settings. Organizations with strict compliance requirements should apply their own input/output filtering on top of the base model. The NVIDIA Nemotron base carries a configurable, enterprise-oriented safety posture.
Pulsar 16B is the right choice for teams that need 30B-class reasoning on hardware or power budgets that cannot accommodate a 30B model. The 43% throughput improvement makes it well suited for high-volume agentic pipelines where cost per token and latency are critical. It fits pipelines already using the Nemotron prompt format. Teams should not choose it if they need vision or audio input without a separate preprocessing step, or if their deployment environment requires a vendor SLA, SOC 2 certification, or audit trail at the inference layer. For managed, compliance-covered inference, AWS Bedrock or Google Vertex with a certified proprietary model is the better option.
Pulsar 16B was not trained from scratch; it is derived from NVIDIA Nemotron 3 Nano 30B through CompactifAI compression. The Nemotron 3 base models were trained on curated public web text and synthetic reasoning traces; the exact training cutoff has not been publicly confirmed by NVIDIA but is estimated to be early-to-mid 2025 based on public release timing. The Apache 2.0 license permits commercial use, modification, and redistribution without royalty obligations. For self-hosted deployments, no data is sent to Multiverse Computing. For CompactifAI API deployments, data governance terms apply per the commercial agreement. No SOC 2 or HIPAA certification specific to Pulsar 16B was announced at the time of writing.
Pulsar 16B is the second major open model release from Multiverse Computing in 2026, following HyperNova 60B, a 50%-compressed version of GPT-OSS-120B released in early 2026. The Pulsar lineage targets the 15-20B active-parameter efficiency tier, while HyperNova targets the 55-65B range. Multiverse Computing's CompactifAI roadmap indicates future Pulsar variants will track NVIDIA Nemotron updates as the Nemotron family evolves. The company was reportedly pursuing a Series C of approximately EUR 500M at a EUR 1.5B valuation in early 2026, with the stated goal of scaling compressed AI across enterprise and edge environments globally.
Pricing
Weights are free to download and self-host under an open-source license. CompactifAI API offers managed token-based access; contact Multiverse Computing for pricing. Costs are reported up to 75% below comparable frontier proprietary models for coding and reasoning workloads.
Key Features
- CompactifAI Compression: Quantum-inspired tensor networks compress the 30B-class Nemotron base into far fewer parameters without retraining, preserving its math and science reasoning scores almost exactly.
- 43% Faster Throughput: FP8 checkpoint on NVIDIA Blackwell delivers 4,808 tokens per second at 32 concurrent requests, a 43% throughput gain over the 30B source model.
- 1 Million Token Context: Inherits the source model's 1,000,000-token context window; needle-in-a-haystack retrieval stays reliable well past the 100K-token mark.
- Three Precision Formats: Ships in BF16, FP8, and NVFP4, scaling from a single consumer GPU up to multi-card server deployments depending on the precision chosen.
- Open Source Licensing: Released under Apache's permissive open-source terms, hosted on Hugging Face under the MultiverseComputingCAI organization, free for commercial use and modification.
- Native Tool Calling: Preserves the Nemotron function-calling schema with structured output and parallel tool-call support, making it a drop-in swap for pipelines already built on Nemotron-class models.
Pros
- Matches the accuracy of its 30B-parameter source model on math and science reasoning benchmarks while cutting active parameters by more than half, a genuine compression-without-degradation result validated on NVIDIA's own hardware.
- Delivers meaningfully higher inference throughput than its source model on the same NVIDIA hardware, cutting cost per token for high-volume workloads.
- Ships as Apache-licensed open weights available in three precision formats (BF16, FP8, NVFP4), deployable anywhere from a single consumer GPU to multi-GPU servers.
Cons
- No vision or audio input; text-only at launch, unlike the broader Nemotron 3 Omni multimodal family it descends from.
- Uses the Nemotron extra_id prompt template rather than the more common ChatML or Llama-3 formats, requiring a pipeline migration for teams switching model families.
- No published SOC 2 certification, vendor SLA, or dedicated Pulsar 16B system card for compliance-heavy deployments.
Benchmarks
- AIME 2025: 87.2% vendor-reported · 23 Jun 2026 — Competition-level maths problems from the 2025 exam, % solved.
- GPQA Diamond: 71.4% vendor-reported · 23 Jun 2026 — PhD-level science questions that are hard to search for, % correct.
- Output speed: 4808 tok/s cited: Artificial Analysis — Median tokens written per second as measured by Artificial Analysis.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
What does Pulsar 16B actually cost?
There is no license fee: the model weights are open and self-hostable, so the only ongoing cost is the GPU hardware or cloud compute running inference. On a single RTX 4090, a full day of agentic coding (roughly 1M tokens in, 200K out) costs well under $1 in hardware time. Multiverse Computing also sells managed token-based access through its CompactifAI API on AWS, Azure, and GCP, priced per contract rather than a published rate card.
Can you use Pulsar 16B without paying?
Yes, entirely: the full model weights are open and free to download, with no usage caps, seat limits, or paid tiers. You pay only for the compute you run it on, whether that is your own GPU or a rented cloud instance. The only paid option is Multiverse Computing's managed CompactifAI API, for teams that would rather not operate their own inference infrastructure.
What are the best alternatives to Pulsar 16B?
The most direct alternative is its own source model, Nemotron 3 Nano 30B A3B, if you have the extra VRAM and want the uncompressed original. For a dense, non-MoE open model in a similar size class with native vision support, Qwen3.8-27B is worth evaluating instead. Both are open-weight and avoid a per-token API bill, unlike closed frontier models.
Pulsar 16B or Nemotron 3 Nano 30B: which should you pick?
Pulsar 16B is Multiverse Computing's compressed version of exactly this model, so the comparison is really compression versus original. On the published reasoning benchmarks, Pulsar 16B tracks the uncompressed Nemotron base closely while using roughly half the total parameters and delivering meaningfully higher throughput on the same hardware. Reach for the full Nemotron A3B build when you need the reference implementation and the widest existing tooling support; reach for Pulsar 16B when GPU memory or inference cost is what's limiting you.
How do you set up Pulsar 16B?
Download the weights from Hugging Face under the MultiverseComputingCAI organization, choosing BF16, FP8, or the smaller NVFP4 checkpoint based on available VRAM. Serve it with vLLM or TensorRT-LLM, and load the Nemotron 3 Nano chat template rather than the ChatML default, since Pulsar 16B was tuned on the extra_id prompt format. Teams that would rather skip infrastructure entirely can reach the same model through Multiverse Computing's CompactifAI API on AWS Marketplace.
Top Alternatives
- Nemotron 3 Nano 30B A3B: Pick Pulsar 16B for cheaper self-hosting on a single GPU; pick Nemotron 3 Nano 30B A3B if you need the uncompressed original for maximum fidelity or broader tooling support today.
- Qwen3.8-27B: Qwen3.8-27B adds vision and video understanding that Pulsar 16B lacks; go with Pulsar 16B instead when math and science reasoning quality matters more than modality coverage.