DeepSeek Open-Sources V4.1-Flash, Opens $0.75 Search API

DeepSeek released an MIT-licensed update to its V4.1-Flash model and opened its Search and Fetch APIs to all developers at $0.75 per 1,000 successful requests, with no change to its core API prices.

By the Numbers

+2 to +11.1 pts
Benchmark gain vs prior V4.1-Flash
7 of 11
Benchmarks still trailing Claude Opus 5
$0.75/1K requests
Search/Fetch API price
MIT
License
DeepSeek Open-Sources V4.1-Flash, Opens $0.75 Search API: key figures
ShareXLinkedInEmail

THE RUNDOWN

1

DeepSeek's own benchmark table shows the new checkpoint beating its predecessor on all 11 published tests, by 2 to 11.1 points, while still trailing Claude Opus 5 on 7 of 11.

2

Opening the Search and Fetch APIs at a fixed per-request price undercuts the usual practice of bundling retrieval behind a subscription or a closed partnership.

3

The release lands as open-weight models, DeepSeek's among them, now process the majority of tokens on major inference gateways like Vercel's.

4

MIT licensing removes a restriction some enterprises cite as a blocker to adopting Chinese open-weight models, even as Beijing weighs new export curbs on the category.

The VC Read

Value Add VC analysis

The benchmark table is the real disclosure here: DeepSeek is publishing exactly where it still loses to Claude Opus 5, which is unusual candor for a model release. For founders building retrieval products, a $0.75-per-1,000-request API from a frontier lab is a pricing floor worth underwriting against before raising a seed round for 'AI search infrastructure.'

Analysis

DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, an MIT-licensed update to the Flash model it launched in September, posting the weights to Hugging Face with no change to its API prices, according to Techi. The company also opened its Search and Fetch APIs to all developers at a flat $0.75 per 1,000 successful requests.

## A Benchmark Table That Admits Where It Loses DeepSeek's own comparison table shows the new checkpoint beating its predecessor on all 11 benchmarks it published, by margins of 2 to 11.1 points -- but the same table shows it still trailing Anthropic's Claude Opus 5 on 7 of those 11 tests. Publishing a loss column is unusual for a model release; most labs lead with the categories they win. Pulse has tracked DeepSeek's broader ambitions, including reports that the company is lining up as much as $12 billion ahead of a potential IPO, so a steady cadence of free, open-weight updates functions as much as a distribution strategy as a research disclosure.

“Publishing a loss column is unusual for a model release; most labs lead with the categories they win.”

## Pricing A Category That Usually Hides Behind Subscriptions Search and retrieval access is typically bundled into a subscription tier or a closed partnership -- OpenAI's and Anthropic's own search features aren't sold as standalone, metered APIs at a published per-request rate. DeepSeek's $0.75-per-1,000-requests price gives any developer a hard number to underwrite against, which matters for the wave of startups pitching "AI search infrastructure" as a venture category: a frontier lab is now the pricing floor, not a future competitor to price against.

The release also lands the same week open-weight models -- DeepSeek's among them -- crossed 71% of token volume on Vercel's AI Gateway, and the same week China floated export curbs on large open-weight models. MIT licensing removes a real adoption blocker for enterprises wary of more restrictive open-weight terms, but it does nothing to resolve the geopolitical risk sitting underneath continued access to Chinese-built weights.

ShareXLinkedInEmail

Key Sources

2 sources
SourceTechi

Reported by Techi · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with The VC Read, a few times a week. Free to subscribe, no spam.

No spam, unsubscribe anytime. Sent via Substack · Privacy · or subscribe on Substack ↗