wbstorm

Compared: Moondream 3 Preview vs 3.1 9B OpenRouter price and HF downloads

On 2026-11-05, Moondream 3 Preview and 3.1 9B share OpenRouter list price and 32k context; official Hub downloads are 39,161 vs 9,160.

You need to line up Moondream 3 Preview and Moondream 3.1 9B A2B on OpenRouter—same list price or not—and see which Hugging Face repos are official. Figures below were read on 2026-11-05 from the public models list and Hub API.

Both OpenRouter ids share 32000 context and identical pricing on the snapshot day, but they are separate routes with different names. Hub weights live under the moondream org; community GGUF uploads are labeled unofficial in the table.

Read both OpenRouter rowsScale per-token USDSplit official vs unofficial Hub

Side-by-side on OpenRouter (2026-11-05)

pricing.prompt is 0.0000003 USD per token and pricing.completion is 0.0000025 USD per token for both ids. Times 1,000,000, that is $0.30 input and $2.50 output per million tokens. context_length is 32,000. OpenRouter created is 1792102385 for preview and 1792102373 for 3.1—both 2026-10-15 UTC. max_completion_tokens is 4,096 for each. architecture modality is text+image→text. is_moderated is false.

32,000
Context (both routes)
$0.30
USD per 1M input tokens
$2.50
USD per 1M output tokens
SourceIdentifierListed / created2026-11-05 figures
OpenRoutermoondream/moondream3-preview2026-10-1532,000 ctx · $0.30 in · $2.50 out / 1M · max out 4,096
OpenRoutermoondream/moondream3.1-9b-a2b2026-10-1532,000 ctx · $0.30 in · $2.50 out / 1M · max out 4,096
Hugging Face (official)moondream/moondream3-previewHub 2025-09-1139,161 downloads · 694 likes · image-text-to-text
Hugging Face (official)moondream/moondream3.1-9B-A2BHub 2026-06-309,160 downloads · 97 likes · image-text-to-text
Hugging Face (unofficial GGUF)SkyContact/moondream3.1-9B-A2B-GGUFHub 2026-10-15460 downloads · 2 likes · community upload

OpenRouter listing day (2026-10-15) does not match every Hub createdAt: the preview weights repo is older on Hub, while 3.1 weights landed 2026-06-30. Download counts rank preview above 3.1 official weights; that is a Hub counter, not a quality score. Do not add the unofficial GGUF row into the official 3.1 download total.

How to re-read both catalogs

  1. 1

    Locate both OpenRouter rows

    Open https://openrouter.ai/api/v1/models and search for moondream/moondream3-preview and moondream/moondream3.1-9b-a2b. Copy context_length, pricing, created, and max_completion_tokens for each.

  2. 2

    Convert USD per token to one million

    0.0000003 × 1,000,000 = $0.30 per million input tokens. 0.0000025 × 1,000,000 = $2.50 per million output tokens.

  3. 3

    Read official Hub repos under moondream/

    Open https://huggingface.co/api/models?search=moondream3 and record downloads and likes for moondream/moondream3-preview and moondream/moondream3.1-9B-A2B.

  4. 4

    Keep unofficial GGUF separate

    SkyContact/moondream3.1-9B-A2B-GGUF is a community GGUF mirror—not the moondream org. Use it only if you self-host quantized weights; it is not merged into official download counts here.

What differs when the price does not

01

Two routes, one price sheet

Preview and 3.1 show the same prompt and completion rates on OpenRouter. Pick the id that matches the weights or docs you follow; the table price alone does not split them.

02

Hub age drives download gaps

Preview’s official repo has been public longer (Hub since 2025-09-11), so downloads exceed 3.1’s official repo (9,160). That ordering is not an eval score.

03

Completion cap vs context

context_length is 32,000 while max_completion_tokens is 4,096 on both OpenRouter entries—plan long outputs accordingly.

To walk a colleague through this table live, open a browser whiteboard at wbstorm. Room setup is in the guide. Related catalog checks: Muse Glimmer 30B price vs downloads and Ling 3.1 Flash Hub vs OpenRouter id.

Fields copied from the APIs

# fetched 2026-11-05
# https://openrouter.ai/api/v1/models
# moondream/moondream3-preview
#   created: 1792102385 = 2026-10-15T22:13:05Z
#   context_length: 32000
#   pricing.prompt: 0.0000003 = 0.30 USD / 1e6 tokens
#   pricing.completion: 0.0000025 = 2.50 USD / 1e6 tokens
#   max_completion_tokens: 4096
# moondream/moondream3.1-9b-a2b
#   created: 1792102373 = 2026-10-15T22:12:53Z
#   context_length: 32000
#   pricing.prompt: 0.0000003
#   pricing.completion: 0.0000025
#   max_completion_tokens: 4096
# https://huggingface.co/api/models/moondream/moondream3-preview
#   downloads=39161 likes=694 createdAt=2025-09-11T19:48:53.000Z
# https://huggingface.co/api/models/moondream/moondream3.1-9B-A2B
#   downloads=9160 likes=97 createdAt=2026-06-30T23:43:59.000Z
# https://huggingface.co/api/models/SkyContact/moondream3.1-9B-A2B-GGUF (unofficial)
#   downloads=460 likes=2 createdAt=2026-10-15T09:50:11.000Z
Are Preview and 3.1 priced differently on OpenRouter?

On 2026-11-05 both ids listed the same pricing.prompt and pricing.completion—$0.30 in and $2.50 out per million tokens—and the same 32,000 context length.

Which Hugging Face repos are official?

Repos under the moondream organization (moondream/moondream3-preview and moondream/moondream3.1-9B-A2B) are treated as official weights here. SkyContact/moondream3.1-9B-A2B-GGUF is a community GGUF upload and is labeled unofficial.

Why do Hub created dates differ from OpenRouter created?

OpenRouter created marks when each id appeared in the OpenRouter catalog (2026-10-15 for both). Hub createdAt is when each repository was published on Hugging Face—preview weights since 2025-09-11, 3.1 weights since 2026-06-30.

Is a higher download count “better”?

No. downloads and likes are Hugging Face counters. Preview’s official repo shows more downloads largely because it has been public longer—not because OpenRouter priced it lower.

Create room