meta
Meta: Muse Glimmer 30B
meta/muse-glimmer-30b
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
Muse Glimmer 30B is Meta's open-weight (Apache 2.0), roughly 30-billion-parameter dense multimodal model, distilled from Muse Spark and built for autonomous agentic work on consumer hardware. Per its model card, it accepts interleaved text and images, emits text, supports a roughly 131K-token context, and uses always-on reasoning with selectable effort levels. Vendor-reported evaluations emphasize tool use, coding, and long-horizon tasks, while the OpenRouter catalog adds limited third-party arena and index measurements. Quantized variants and a speculative-decoding drafter target 24-32 GB memory envelopes.[1][2]
Strengths
- Purpose-built for end-to-end agentic workflows: the model card reports reliable schema-based tool calling, multi-step planning over long horizons, and recovery when tool calls fail.[2]
- A dedicated perception encoder accepts interleaved text and images (screenshots, charts, documents) within a roughly 131K-token context, per the catalog listing and model card.[1][2]
- Designed for local deployment: 4-bit quantized builds and a companion drafter for block speculative decoding target consumer GPUs, with the vendor reporting minimal measured quality loss.[2]
Limitations
- Evidence is mostly vendor-published plus a catalog listing; independent verification is thin, so benchmark claims and speed figures should be validated against your own workloads.[1][2]
- The model card warns it can produce inaccurate, biased, or objectionable output, may err in multi-step reasoning on novel scenarios, processes video only as frames, and was not evaluated on all trained languages.[2]
- Reasoning is mandatory rather than optional per the catalog, and the maker recommends deployment guardrails (e.g., confirmation for irreversible actions) instead of standalone use.[1][2]
Where to use it
- Local coding and task-completion agents that plan, invoke tools, debug, and recover from errors without cloud infrastructure, as suggested by the provider's intended-use list.[2]
- Multimodal assistants that interpret screenshots, charts, and documents alongside conversation in agentic or information-rich environments.[2]
- Synthetic data generation or LLM-as-a-judge evaluation for downstream model development, suggested as a fit by the model card.[2]
Sources
- [1] OpenRouter catalog · Meta: Muse Glimmer 30BSources checked: Nov 13, 2026
- [2] Model documentation · Meta: Muse Glimmer 30BSources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
Muse Glimmer 30B is a dense, open-weight multimodal model from Meta Superintelligence Labs, distilled from Muse Spark and optimized for autonomous agents on consumer hardware. It is suited for long-horizon...
Specifications
- Model ID
- meta/muse-glimmer-30b
- Canonical ID
- meta/muse-glimmer-30b-20260810
- Creator
- meta
- Added to OpenRouter
- 09.08.2026
- Context length
- 131,072
- Modality
- text+image->text
- Input modalities
- text, image
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.300 / out $1.2
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | CC — below the B band | reasoning: mandatory · default medium · efforts: xhigh, high, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Coding | BB — next 30% of the cohort | — |
| Agents | BB — next 30% of the cohort | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · default medium · efforts: xhigh, high, medium, low · parameters: include_reasoning, reasoning, reasoning_effort |
| Document analysis | No measurement | — |
| Vision | No measurement | input: image |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Intelligence index | 6.3 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Coding index | 49 | BB — next 30% of the cohort | 13.11.2026 |
| Artificial Analysis | Agentic index | 8.5 | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · 3dwin rate 26% · rank 100 | 1,070Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 35.7% · rank 81 | 1,160Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · gamedevwin rate 29.9% · rank 94 | 1,109Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 39.6% · rank 63 | 1,198Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · websitewin rate 39.1% · rank 73 | 1,186Elo | BB — next 30% of the cohort | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| DeepInfrabf16 | 131,072 | $0.300 | $1.2 | — | — | 100% |
Provider data retrieved: 13.11.2026