Moondream 3.1: 9B MoE vision model with 2B active parameters, 32K context, $0.30/$2.50 per 1M tokens
Moondream has released Moondream 3.1, a mixture-of-experts vision-language model with 9B total and 2B active parameters and a 32K-token context window. It is listed on OpenRouter at $0.30 per 1M input tokens and $2.50 per 1M output tokens, with an optional reasoning mode.
Moondream 3.1 — Quick Specs
Moondream has released Moondream 3.1, a vision-language model with 9B total parameters and 2B active per token, built as a mixture-of-experts (MoE). It has a 32K-token context window and is listed on OpenRouter at $0.30 per 1M input tokens and $2.50 per 1M output tokens. The listing gives a release date of October 15, 2026.
Specifications
According to the OpenRouter listing, Moondream 3.1 is the successor to Moondream 3 Preview.
| Spec | Detail |
|---|---|
| Model ID | moondream/moondream3.1-9b-a2b |
| Architecture | Mixture-of-experts |
| Total parameters | 9B |
| Active parameters | 2B |
| Context window | 32K tokens |
| Input price | $0.30 per 1M tokens |
| Output price | $2.50 per 1M tokens |
| Release date | October 15, 2026 |
| Reasoning mode | Optional |
The listing says the model handles visual question answering, captioning, OCR, and visual reasoning over images. Model weights are linked from the listing.
Availability and performance
The only provider currently listed on OpenRouter is Prime Intellect. It serves the model at the prices above, with a P50 latency of 0.28 seconds. Throughput and uptime figures are not yet populated. OpenRouter also reports not enough usage data from apps to show adoption.
What is not disclosed
The listing does not include:
- Benchmark scores for Moondream 3.1
- A training data cutoff date
- A list of changes relative to Moondream 3 Preview
- Details on how the reasoning mode is enabled or how it affects pricing and latency
The description of Moondream 3.1 on OpenRouter is word-for-word the same as that of Moondream 3 Preview, and both are priced at $0.30 / $2.50 per 1M tokens with a 32K context. Any quality improvement over the Preview is therefore unverified. The source material gives no basis for comparison.
What this means
The notable choice here is the architecture. With 2B of 9B parameters active, per-token compute should be close to that of a small dense model, while total capacity is larger. That is the usual rationale for MoE in latency-sensitive vision workloads, and the 0.28s P50 latency fits that profile, though the figure comes from a single provider and a single listing.
Pricing is more ambiguous. At $0.30 input and $2.50 output, the output price is over 8x the input price. For captioning, OCR and visual QA, where outputs are often short, this may be manageable. If the optional reasoning mode generates long outputs, costs could rise quickly. Teams should measure this before committing to it.
The unchanged price and specifications relative to Moondream 3 Preview suggest 3.1 is a refinement rather than a structural change, but Moondream has not said so in the material available. Builders considering a migration from the Preview should run their own evaluations on OCR and visual reasoning tasks until the company publishes benchmark results and release notes.
Related Articles
Moondream 3 Preview lists on OpenRouter: 9B MoE vision model, 2B active, 32K context, $0.30/$2.50
Moondream 3 Preview is a vision-language model with 9B total and 2B active parameters (mixture-of-experts), a 32K-token context window and an optional reasoning mode. OpenRouter lists it at $0.30 per 1M input tokens and $2.50 per 1M output tokens, served by Prime Intellect.
StepFun releases Step 5 Preview: 600B MoE with 1M context at $1/$2.70 per 1M tokens
StepFun has listed Step 5 Preview, a sparse Mixture-of-Experts model with 600B total and 27B active parameters and a 1.0M-token context window. It is priced at $1 input and $2.70 output per 1M tokens on OpenRouter. StepFun positions it as its flagship model for agentic work.
Nvidia releases Nemotron 3.5 Super VL: open 120B multimodal MoE with 12B active parameters
Nvidia has released Nemotron 3.5 Super VL, an open multimodal Mixture-of-Experts model with 120 billion total and 12 billion active parameters, built for continuously running enterprise agents. Two compact Bolt models for local PCs and edge devices launch alongside it, though the 4 GB variant ships later this month.
Unbiased releases Pareto 26.9: 262K-context composite model at $2.50/$7.50 per 1M tokens
Unbiased has released Pareto 26.9, a multimodal composite model with a 262K-token context window, priced at $2.50 per 1M input tokens and $7.50 per 1M output tokens. It is listed on OpenRouter with a Oct 13, 2026 release date. A cheaper 1.0M-context preview, Pareto 26.10, is already listed alongside it.
Comments
Loading...