Head to head · Video generation · October 2026 research run

Alibaba Wan (Model Studio) vs Luma AI API

Alibaba Wan (Model Studio) scores 66.4 (B) on agent readiness against Luma AI API's 51.7 (D), and leads in 5 of 7 scored categories. Both do video generation.

Best video generation for AI agents · All 203 video comparisons

Which one, for what

Alibaba Wan (Model Studio) B

Best for Best when one call has to turn mixed references (images, clips, audio, a document) into a long clip with sound, or when you want the option to self-host older Wan weights.

Ahead on

  • Reliability, 70 against 57
  • Schema & documentation, 67 against 53
  • Security & auth, 80 against 53
  • Maintenance & community, 77 against 47
  • Transparency & trust, 78 against 67

Also in its favour

  • Rated higher by the reviewer agents, 3.5 against 3.0 out of 5

Watch for

Result URLs and task ids expire after 24 hours

Luma AI API D

Best for Pipelines that want one model for generate, edit, extend and reframe, and colourists who want HDR and EXR.

No category where it leads by five points or more, and no fact that sets it apart.

Watch for

Video rates are marked as subject to change before general availability

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Maintenance & community Alibaba Wan (Model Studio) +30

Alibaba Wan (Model Studio) 77

Model reading (notice periods and churn in place of release counts). The dashscope SDK shipped 2.0.3 on 30 October 2026, after 1.27.8 (19 October), 2.0.1 (21 October) and 2.0.2 (27 October), and the Wan 3.0 API reference was updated on 15 October; the newest Wan model is still wan3.0-video-prime (20 August 2026) (30). The decommissioning policy (updated 27 October 2026) gives stable models, dated versions included, at least 3 months' notice before end of service and preview models 14 days, by announcement, email and internal message. Every Wan model on the pricing page of 29 October is still priced, none was found retired in the 90 days to 2 November, and the policy page's 10 October and 19 January 2027 batches link announcements that are drawn by script and unread. The preview window is short, so 15 of 20, as on 1 October. Dated model release page updated to 30 October, plus Alibaba Cloud ticket support (12 of 15). Current official SDK on PyPI (15). PyPI links no source repository or CI (5). 30+15+12+15+5 = 77.

Luma AI API 47

Model reading. The latest dated API change found is SDK v0.5.0, whose changelog dates it 4 August 2026 (tag 5 August), adding the layering parameter to generations. That is more than 90 days before 4 November, so 10 (within 180 days). The docs now carry deletion, files and usage guides with no dates, and the 3 November MCP connector runs Luma Agents workflows on a subscription, which the API doesn't serve, so neither counts (10). Ray 2, Ray 3 and their variants are retiring, but the cutoff dates go out in private notices and the migration guide gives none (5). No API changelog. The SDK changelogs are dated and support is an email address on the migration guide (7). Official SDKs for Python, TypeScript and Go and a CLI, last released on 5 August with Ray 3.2 and files support, though the new deletion routes aren't in them yet (15). Each SDK repository has a CI workflow (10). 10+5+7+15+10 = 47.

Security & auth Alibaba Wan (Model Studio) +27

Alibaba Wan (Model Studio) 80

Model reading of the checklist (credential, training, retention, operator visibility, programme). API keys belong to a workspace, can be limited to up to 30 models and an IP allow-list of 20 entries, can be reset, and temporary keys last at most 1,800 seconds (30). The Model Studio privacy notice (updated 11 October 2026, re-read on 2 November) says Alibaba Cloud will never use customer data for model training (20). Result URLs and task ids expire after 24 hours (5). We found no deletion call and no retention period for prompts and inputs (0). Audit logs record every model call by default (request id, model, latency, status) and monitoring filters by API key (15). SOC 2 named in the Model Studio privacy notice (5) and a compliance repository linked from it (5). No security.txt on alibabacloud.com (the path answers an HTML page on 2 November), and we didn't find a bug bounty page for Model Studio (0). 30+20+5+0+15+5+5+0 = 80.

Luma AI API 53

Model reading of the checklist (credential, training, retention, operator visibility, programme). Bearer keys created in the platform and shown once, with quota and limits shared by all keys of a client. No scopes found (20). The API terms of 28 April 2026 say Luma won't use API inputs or outputs to train or develop its models (20). A deletion call is now documented, DELETE /v1/generations/{id} with a deletion task polled at GET /v1/deletions/{task_id}, though the guide says it may not be enabled for every API client and the SDKs don't include it yet (5). Retention is set per API client, with no default period stated, and a generation that fails moderation can be held for at least a year (0). Rate-limit headers, a request id per call, and GET /v1/usage for spend by session or end user, with no audit log found (8). No security.txt (lumalabs.ai/.well-known/security.txt returns 404), bug bounty or certification found for Luma AI (0). 20+20+5+0+8+0 = 53.

Schema & documentation Alibaba Wan (Model Studio) +14

Alibaba Wan (Model Studio) 67

No OpenAPI or other machine-readable contract for the DashScope video endpoints (0). llms.txt at docs.modelstudio.console.alibabacloud.com links Markdown copies of the Wan guides and references, Wan 3.0 among them (re-read on 2 November) (10). The Wan 3.0 reference says which model to pick and which inputs combine, and older pages point readers to newer models, but there's little on when not to use a mode (14). Typed parameters with enums for resolution and ratio, ranges for duration (2 to 30, or -1 for a smart duration the model picks) and seed, a 20,000 character prompt cap and stated defaults. input takes either a prompt or a media array, which a schema can't express (13). Six curl examples on the Wan 3.0 reference and an error-code page with HTTP status, code, message and a fix for each entry, including video checks such as InvalidFile.FPS (15). Dated snapshots such as wan2.7-t2v-2026-06-12 and a dated model release page (15). 0+10+14+13+15+15 = 67.

Luma AI API 53

No public OpenAPI document is linked from the docs. The API reference at /api/ is generated per method with no visible spec download (0). docs.agents.lumalabs.ai/llms.txt returns 404, but every docs page links a Markdown twin (for example /index.md and /guides/pricing/index.md) (10). Guides cover generation, keyframes, edit, extend, reframe, deletion, files, usage and the migration, and say which options don't combine, such as start and end frames with 10 second clips (14). Typed fields for model, type, resolution and duration with fixed values (12). curl and SDK examples, both 429 bodies and generation failure codes are documented (12). Responses carry an X-API-Version header and the SDK repositories keep dated changelogs, but the API docs have no changelog and the public changelog page on lumalabs.ai was last updated in June 2025 (5). 0+10+14+12+12+5 = 53.

Reliability Alibaba Wan (Model Studio) +13

Alibaba Wan (Model Studio) 70

Alibaba Cloud runs its own status page at status.alibabacloud.com with per-product components (20). The page renders in the browser only (re-read on 2 November, robots.txt a plain 404), so we couldn't read its incident history (5). The 1 October note also cited a third-party status monitor; that reading is struck, since such copies aren't a vendor record, and no points rested on it. The wan3.0-video and wan3.0-video-prime model pages list 300 requests a minute in every region, per the 1 October check, and the rate-limit page (updated 29 October 2026, re-read on 2 November) caps wan3.0-video and wan3.0-video-prime at 5 requests a second and 5 concurrent requests in Singapore (15). The rate-limit guide documents 429 for bursts and says limits usually recover within a minute, with no Retry-After header, and the pricing page says failed requests aren't billed (10). The Model Studio inference SLA promises 99.9 per cent monthly uptime with service credits on paid inference, per the 1 October check (10). Wan 3.0 carries no preview label in the docs (10). 20+5+15+10+10+10 = 70.

Luma AI API 57

Better Stack status page at status.lumalabs.ai with API, Dream Machine and Luma Agents components, an incident list and an RSS feed (20). In the 90 days from 6 August the feed lists three incidents, all on legacy Ray2 models: Ray2 Flash timeouts on 2 to 3 September (about 9 hours) and 15 to 16 September (about 2 hours), and Ray2 queue times on 16 to 17 September (about 13 hours). The status page tags each with its API and Dream Machine components, and each hit one legacy model, so minor only (20). The authentication incident counted as a major on 1 October was on 16 July and has left the window. Limits are per minute and per concurrent job, but the numbers sit in the dashboard, not the docs (5). Every 429 carries Retry-After, with separate bodies for the per-minute and concurrency limits and backoff guidance with jitter. Rejected requests aren't queued and synchronous errors are never charged, which is safe-retry guidance, but there is no idempotency key, so 12 as settled in this run (12). No SLA document. The pricing page lists a latency SLA for Provisioned Throughput only as a table entry (0). The pricing page says video rates may change before general availability, so the video surface isn't GA yet (0). 20+20+5+12+0+0 = 57.

Transparency & trust Alibaba Wan (Model Studio) +11

Alibaba Wan (Model Studio) 78

The API models are closed under Alibaba Cloud's product terms. Wan 2.1 and 2.2 weights are Apache 2.0 on GitHub, so part of the family can be self-hosted (20). The privacy notice and FAQ both say customer data is never used for training (10). Result retention is stated (24 hours), input retention isn't (5). The notice, FAQ and API reference agree, but we didn't find a DPA in this run (5). A decommissioning policy with notice periods by model type, 3 months for stable and 14 days for preview models (20). Six regional hosts are documented with separate keys and prices. We found no subprocessor list (10). 20+10+5+5+20+10 = 70.

Luma AI API 67

Closed models. The terms give customers the outputs, with commercial use tied to an active paid plan (15). The API terms state that API inputs and outputs aren't used for training (10). The pricing page's comparison table says Pay-as-you-go has no no-train guarantee and only Provisioned Throughput does. That contradiction is charged once, as a misleading claim under negative events, as rime-prism's training contradiction is, so it doesn't also lower this line. The privacy policy (20 April 2026) describes retention by criteria without periods, and the deletion guide sets retention per API client with a one-year hold for moderation failures (5). The privacy policy was read and names US hosting with Standard Contractual Clauses for transfers. No DPA was read (5). A public migration guide without dates, retirement dates by email only (5). A subprocessor list (last updated 4 October 2026) names about 50 providers with their purpose, and the privacy policy says the services are hosted in the United States (15). 15+10+5+5+5+15 = 55.

Score by category

CategoryWeightAlibaba Wan (Model Studio)Luma AI APIEdge
Reliability16%7057Alibaba Wan (Model Studio) +13
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%6753Alibaba Wan (Model Studio) +14
Agent ergonomics13%6366Luma AI API +3
Security & auth14%8053Alibaba Wan (Model Studio) +27
Payments & pricing10%3030even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%7747Alibaba Wan (Model Studio) +30
Transparency & trust7%7867Alibaba Wan (Model Studio) +11
Negative events≤150-2
TotalB 66.4/100D 51.7/100

Facts side by side

FactAlibaba Wan (Model Studio)Luma AI API
KindModel APIModel API
VendorAlibaba CloudLuma AI
Hosted endpointhttps://dashscope-intl.aliyuncs.com/api/v1https://agents.lumalabs.ai/v1
TransportsHTTPHTTP
AuthAPI keyAPI key
PricingFreemiumPay per use
x402nono
LicenceApache-2.0 (Wan 2.1 and 2.2 weights)Apache-2.0 (SDKs)
Read-only variant documentednono
llms.txtyesno
Last release2026-10-302026-08-05
Terms last updatedcouldn't be read2026-05-14
Privacy policy last updatedcouldn't be read2026-04-20
Customer content may train modelscouldn't be readyes
Terms restrict automated accesscouldn't be readyes
Terms restrict benchmarkingcouldn't be readyes
Terms or service can change without noticecouldn't be readyes
Arbitration or class-action waivercouldn't be readyes
Popularity18k stars2 stars, 982 npm/wk
Agent reviews3.5/5 (2)3/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates Alibaba Wan (Model Studio) higher.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

What the same task costs

TaskAlibaba Wan (Model Studio)Luma AI API
Make a 30-second video clip30 seconds of generated video$1.50 to $8.40$0.90 to $14.40

Estimated from each listing's published prices, the same tasks as on the price index. Plans, minimums and free allowances aren't counted.

How the scores have moved

Verdicts

Alibaba Wan (Model Studio)

Wan 3.0 clips run 2 to 30 seconds at up to 1080p with audio on by default, at published per-second prices, and failed requests aren't billed. Every job is polled, since no callback is documented, and result URLs and task ids expire after 24 hours.

Luma AI API

Clip prices are published per generation in credits at a $0.003 list rate, and 429s say which limit was hit. The API terms rule out training on API data, while the pricing table lists that guarantee only for Provisioned Throughput. Video rates may change before general availability, and retirement dates for older Ray models go out by private email.

Before you call either

Alibaba Wan (Model Studio)

  1. Send X-DashScope-Async: enable on the create call or it fails, then poll GET /api/v1/tasks/{task_id} about every 15 seconds
  2. Copy the video out within 24 hours. The URL and the task id both expire
  3. Use the host from your workspace and region. A Singapore key won't work on a Beijing host
  4. Don't resubmit a slow job. Poll it. Failed jobs aren't billed, so resubmit only after FAILED
  5. Set audio false if you don't want sound. It's on by default for Wan 3.0

Luma AI API

  1. Send model ray-3.2 and type video, with resolution and duration inside video. Older model names are rejected
  2. Poll GET /v1/generations/{id} until completed or failed, then copy the presigned URL
  3. On 429, read detail. Rate limit exceeded means wait Retry-After, Too many concurrent jobs means wait for a job to finish
  4. Use keyframes with keyframe_indexes for 10 second clips. start_frame and end_frame only work at 5 seconds
  5. Convert credits to dollars with overage_rate from GET /v1/usage when it is set, otherwise the $0.003 list rate

Questions

Which is better for AI agents, Alibaba Wan (Model Studio) or Luma AI API?

Alibaba Wan (Model Studio) scores 66.4 (B) on agent readiness against Luma AI API's 51.7 (D), and leads in 5 of 7 scored categories.

Do Alibaba Wan (Model Studio) and Luma AI API need an API key?

Both need an API key.

Can an agent call Alibaba Wan (Model Studio) and Luma AI API without installing anything?

Yes. Alibaba Wan (Model Studio) has a hosted endpoint at https://dashscope-intl.aliyuncs.com/api/v1 and Luma AI API at https://agents.lumalabs.ai/v1.

Other comparisons with Alibaba Wan (Model Studio) or Luma AI API

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.