HyperNova 60B
Released
2026-01-02
Last refreshed
2026-10-05
Status
Researched 1d ago
Open sourceCommercial use: permittedCodingRAGAgentsLong contextClassificationJSON / Tool use
HyperNova 60B is available now for coding, rag, and agents with open-source and 131k context; evaluate it while provider pricing coverage matures.
Multiverse Computing releases · 1 in the last 12 monthsChangelog →
Specifications
- Family
- HyperNova
- Released
- 2026-01-02
- Context
- 131k
- Parameters
- 60B (4.8B active)
- Architecture
- Mixture of Experts
- Specialization
- general
- Openness
- Open source
- License
- Apache 2.0OSI-approvedCommercial use: permitted
- Weights
- Available
- Code
- Unknown
- Training
- Fine-tuned
Created by
Pricing
No tracked provider token pricing is available yet.
Links
About
HyperNova 60B is Multiverse Computing's compressed version of OpenAI's gpt-oss-120b. CompactifAI tensor-network compression cuts the 117B MoE to about 59–60B total parameters (4.8B active, 80 experts with 4 routed per token; MXFP4, ~32 GB), so it runs on a single 40 GB GPU. Capability lost to compression is then recovered by knowledge distillation. Versions: the original HyperNova-60B (2 Jan 2026); 2602 (Feb 2026, adds tool calling); 2605 (May 2026, better coding and general scores). It keeps gpt-oss's harmony format, configurable reasoning effort and OpenAI-style function calling. Text-only and English-centric. Weights are Apache 2.0.
Provider price ladder
No tracked provider token pricing is available for this model yet.
Capabilities
ReasoningJSON / Tool useStructured Outputs
Benchmark peer barsfor Coding
LiveCodeBenchRank 46 of 58
Aider PolyglotRank 27 of 34
Benchmark scores(7)
Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| Humanity's Last Exam | 15.0 | HLE (accuracy)Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: NeMo-Skills, 1 repeat, judge openai/gpt-4o Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). Tools not stated. | Source |
| MMLU PRO | 76.8 | MMLU-Pro (accuracy)Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: NeMo-Skills, 1 repeat Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). | Source |
| AIME 2025 | 90.0 | AIME 2025 (accuracy)Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: NeMo-Skills, 10 repeats Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). | Source |
| Google-Proof Q&A | 71.9 | diamondObserved 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: NeMo-Skills, 5 repeats Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). | Source |
| LiveCodeBench | 68.7 | LiveCodeBench v5 (test_v5_2407_2412, Jul–Dec 2024)Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: NeMo-Skills, 3 repeats Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). Older v5 split, not v6; not comparable with v6 rows. | Source |
| τ-bench | 61.7 | τ²-Bench Telecom (accuracy)Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: EvalScope 1.4.1, 3 repeats, max steps 100 Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). | Source |
| Aider Polyglot | 34.2 | Aider polyglot, whole edit format, 225 exercises, --tries=2Observed 2026-08-07 | Configuration: HyperNova 60B 2605 weights; reasoning effort high; vLLM 0.13.0; temp 1.0, top_p 1.0 Harness: Aider-AI/aider, 2 repeats Evaluator: Multiverse Computing (vendor-reported) Notes: First-party model card table 'HyperNova 60B Benchmark Comparison' (card last modified 2026-08-07); same numbers in Multiverse article 'HyperNova 60B v2605' section (updated 2026-06-17). | Source |
Migration checks
No linked migration route is available for this model yet.
API versions
MultiverseComputingCAI/HyperNova-60B (2026-01-02)MultiverseComputingCAI/Hypernova-60B-2602 (HF 2026-02-20)MultiverseComputingCAI/Hypernova-60B-2605 (HF 2026-04-30; API upgrade 2026-05-06)hypernova-60b (CompactifAI API id, deprecated 2026-09-09)Rankings & picks(1)
Created by
Pricing
No tracked provider token pricing is available yet.
Links