DeepSeek Slashes Free File Attachment Limits Silently
Key insights
- DeepSeek reduced free-tier file attachment limits from 50 files at 100 MB each to an undisclosed lower cap with zero public notice.
- The restriction follows Dreamina tripling paid video generation costs overnight, suggesting rising compute pressure across Chinese AI providers.
- Power users running multi-document analysis workflows on DeepSeek's expert model are the most directly impacted segment.
Why this matters
Practitioners building document-heavy workflows on top of DeepSeek's free tier now face an invisible, unannounced constraint that can silently break production pipelines with no warning or documented fallback. The pattern of Chinese AI providers restricting access without changelog entries creates operational risk for any team that adopted these models specifically because of their generous free limits, meaning technical due diligence now needs to account for undisclosed policy changes as a failure mode. For founders evaluating Chinese AI APIs as cost-effective infrastructure, this signals that free-tier economics that drove adoption decisions may not be stable or even reliably communicated.
Summary
DeepSeek quietly dropped file attachment limits for free users of its flagship expert model, cutting what was previously a generous 50-file, 100 MB-per-file cap per conversation to an undisclosed lower threshold with no changelog, no announcement, and no advance notice.
The move lands one week after Dreamina tripled video generation costs overnight for paid subscribers, pointing to a broader pattern among Chinese AI providers tightening free-tier access as compute costs climb. DeepSeek has not publicly stated what the new limits are, leaving users to discover the restriction through failed uploads.
Essentially: (DeepSeek, Dreamina) are both pulling back free-tier generosity with no public communication.
- Previous limit: 50 files at 100 MB each per conversation, now replaced by an unannounced lower cap.
- Affected workflow: multi-document analysis, the primary use case driving power-user adoption of DeepSeek's expert model.
- No changelog or support notice has been published as of reporting.
As frontier model costs rise, the race-to-the-bottom on free-tier access that helped Chinese AI providers gain Western users is quietly reversing.
Potential risks and opportunities
Risks
- Teams with automated document-ingestion pipelines built on DeepSeek's free tier could experience silent failures, with no error message clearly attributing the break to the new attachment cap.
- Enterprises that adopted DeepSeek as a cost-saving alternative to OpenAI or Anthropic for document workflows may face unplanned migration costs if limits continue tightening without notice.
- DeepSeek's credibility with Western power users erodes further if the new cap remains undisclosed for weeks, accelerating churn toward providers with published and stable API terms.
Opportunities
- API-first document analysis providers (LlamaParse, Unstructured.io) gain a direct acquisition pitch for DeepSeek users now hitting undisclosed limits in active workflows.
- OpenAI and Anthropic can capitalize by prominently publishing and stabilizing their own file and context limits, positioning transparent API terms as a competitive differentiator.
- Workflow orchestration platforms (Zapier, Make, Relevance AI) that offer model-agnostic routing could market automatic failover to alternative models when provider limits are hit as a reliability feature.
What we don't know yet
- What the new file attachment cap actually is: DeepSeek has not published the figure as of May 15, 2026.
- Whether paid DeepSeek tiers are subject to the same reduction or whether this is exclusively a free-tier rollback.
- Whether DeepSeek plans to retroactively document the change or communicate limits going forward, given no support notice has appeared.
Originally reported by reddit.com
Read the original article →Original headline: DeepSeek Silently Cuts Free-Tier File Attachment Limits From 50 Files / 100 MB Per Conversation to Undisclosed Lower Cap With No Notice