Models · Meta (Meta Superintelligence Labs, Muse) · Out sinceReleased 10 Aug 2026
Muse Glimmer 30B#36 for research and analysis.#36 for research and analysis, best at high effort.
Muse Glimmer 30B is made by Meta (Meta Superintelligence Labs, Muse). Among the models we track it ranks #36 for research and analysis. It's cheap to use. Its makers have published it, so you can run it on your own servers.
Open-weights (Apache 2.0) 30B local-agent model distilled from Muse Spark; controllable reasoning effort. Context from OpenRouter listing. No Meta list API price (OpenRouter $0.35/$1.50).
- —
- 26.2 / 100 · #36
- —
- Cheap$0.35 / $1.50
- $0.64
- 131K
- —
- 35 (13 independent13 indep.)
- 10 Aug 2026
- Meta (Meta Superintelligence Labs, Muse)
- text, image
Scores are out of 100 for each area, compared with every model we track; “#” is its rank. Context is how much text it can read at once (1M is roughly 700,000 words). Blended price mixes the cost of what you send and what it writes back.
You can't choose how long this one thinks.
No adjustable reasoning setting listed.
OpenRouter is a service that gives access to many AI models in one place. This model is listed there as meta/muse-glimmer-30b.
Route it as meta/muse-glimmer-30b at $0.35 in / $1.50 out per 1M tokens, 131K context. Listed since 9 Aug 2026.
frequency_penaltyinclude_reasoninglogit_biasmax_tokensmin_ppresence_penaltyreasoningreasoning_effortrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_p
Run it yourself
Self-hosting
On your own servers,Licence, sizeunder your own control.and hardware.
- Yes
- 29.6B
- No
Runs on a powerful laptop or workstation.
≤ 24 GB at 4-bit (one consumer GPU / 32 GB Mac). Weights ≈ 31 GB at 8-bit, 16 GB at 4-bit (+10–30% for KV cache).
Languages: Languages: Card: trained on data from 100+ languages (none named).
Training it further: Fine-tuning: Card: BF16 weights released 'for fine-tuning and research'; Unsloth 'Fine-tune Muse Glimmer' guide. No pretrained base (distilled from Muse Spark).
Quantisations: bf16, gguf 4-bit x2 (official: meta-models/Muse-Glimmer-30B-GGUF, K-Quant-Dynamic 32GB / K-Quant-17GB 24GB), executorch pte (official), nvfp4 (NVIDIA), fp8-block (community: RedHatAI). Engines: Transformers, llama.cpp (build b10353+), ExecuTorch.
Download (Hugging Face)Weights on Hugging Face ↗huggingface.co ↗huggingface.co ↗unsloth.ai ↗
Every test result
Every result
All the numbers,with where they came from.
Every result we've found for this model, with who measured it and a link to where we read it.
All raw rows (vendor and independent kept separate), with setting label, source, date, harness and notes.
| NotesNotes | ||||||
|---|---|---|---|---|---|---|
| AA-Briefcase v1.1 | 474 | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | |
| AA-LCR | 83.3% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-LCR accuracy, AA-run (long multi-document reasoning) |
| AA-LCR | 80.0% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| AA-Omniscience Accuracy | 27.0% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | |
| AA-Omniscience Hallucination Rate | 81.9% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | share of non-correct answers that were wrong instead of abstaining; = 1 - AA 'omniscienceNonHallucination' field (equals breakdown.hallucinationRate on the eval page) |
| AA-Omniscience Index | -32.9 | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA-Omniscience Index (-100..100): correct minus incorrect, abstentions not penalised |
| AIME 2026 | 94.7% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| Artificial Analysis Intelligence Index | 17.5 | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer |
| Artificial Analysis output speed | 146 tok/s | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer; median output tokens/sec across AA-tracked API providers (not a self-hosted measurement) |
| BEAM 128K | 65.1% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| CharXiv Reasoning (no tools) | 78.8% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| DeepSearchQA | 74.6% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| EQ-Bench Creative Writing v3 (Elo) | 1798 | Defaultmeta-models/Muse-Glimmer-30B | Independent testIndependentEQ-Bench ↗ | 1 Oct 2026 | — | leaderboard rank 26; rubric score 16.26/20; slop 12.35; avg length 5837 chars; Elo judged by Claude Sonnet 4.6; effort not stated by EQ-Bench |
| GAIA2 | 43.3% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| GDPval-AA (v1) | 953 | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | Listed as 'GDPval-AA' (version not stated). |
| GDPval-AA v2.1 | 774 | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | 95% CI 746.44-802.05 |
| GPQA Diamond | 83.5% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer |
| GPQA Diamond | 83.5% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| Humanity's Last Exam | 22.0% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer |
| Humanity's Last Exam | 22.0% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | Text, no tools. |
| IFBench (AllenAI) | 77.0% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| MCP Atlas | 75.5% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| MMMU-Pro (no tools) | 74.0% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| OmniDocBench v1.5 | 75.8% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| OSWorld-Verified | 65.9% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| SciCode | 44.9% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer |
| SciCode | 43.6% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| ScreenSpot-Pro | 75.4% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| SkillsBench | 44.3% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | With skills. |
| SWE-Bench Pro (public, v1) | 51.2% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| SWE-bench Verified | 76.0% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| Terminal-Bench 2.1 | 51.7% | HighMuse Glimmer (High) | Independent testIndependentArtificial Analysis ↗ | 1 Oct 2026 | — | AA slug muse-glimmer |
| Terminal-Bench 2.1 | 51.7% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| WildClawBench | 47.6% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — | |
| τ³-bench Banking | 23.5% | HighHigh reasoning | Maker's own figureVendor-reportedMeta ↗ | 10 Aug 2026 | — |