Muse Glimmer 30B.
A dense 30B open-weights model from Meta that takes text and images and is tuned for local agent work, with a 4-bit build that fits in 24 to 32 GB of VRAM.
1× Nvidia Q RTX 6000 · $0.13/hr.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
Frequently asked.
How do I run Muse Glimmer 30B?
Where can I access Muse Glimmer 30B?
How much does it cost to run Muse Glimmer 30B?
Is Muse Glimmer 30B open-source or proprietary?
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/小时为当前最低报价。多卡配置在有同规格实例出租时使用真实实例价格,否则为单卡价格乘以卡数。
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
96 GB |
$0.50/hr
估算:4 × 单卡价格
Vast.ai · 按需
|
— | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
48 GB |
$0.25/hr
估算:2 × 单卡价格
Vast.ai · 按需
|
— | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
24 GB |
$0.13/hr
Vast.ai · 按需
|
31.4 | Compare → |
What it costs per month across providers.
Estimate your monthly bill for Muse Glimmer 30B across every host that publishes per-token pricing. Slide your token volumes; the chart + table re-rank cheapest-first.
最便宜的服务商在左侧。
每月总成本 — 输入与输出 token 合计。
账单明细。
租 GPU,而不是按 token 付费。
对于 Muse Glimmer 30B 这类开放权重模型,你可以租一块 GPU 自己提供推理服务。算法:最便宜的 GPU 租价 × 每月 730 小时。电费已包含在租用的每小时价格中。
假设 GPU 全天 24 小时运行,利用率约 85%。如果流量时高时低,API 会更省钱,GPU 则可能更贵(空闲时段仍要付租金)。盈亏平衡分析见自托管与 API 盈亏平衡工具。
About Muse Glimmer 30B.
Muse Glimmer is a dense 30B-class model (about 29.6B parameters including its vision encoder) from Meta Superintelligence Lab, released with open weights under Apache 2.0. It reads text and images, outputs text, and handles over 131,000 tokens of context. It is built for agent tasks that run locally: tool calls, multi-step reasoning, recovering from failed calls and reading screenshots or documents. Meta compresses it to roughly 4-bit so it fits in 24 to 32 GB of VRAM, and ships a small drafter model for faster generation.