Call Center Doctors tests DeepSeek V4.1 Flash on Nvidia H200s, finds costs higher than claimed
The Call Center Doctors, a call center consultancy, rented four Nvidia H200 GPUs on September 27 to test DeepSeek V4.1 Flash against Anthropic’s Claude Opus 5.5. The goal was to verify claims that DeepSeek is 80x cheaper, but the consultancy found the on-demand costs topped $440.88 per day—more than double the $184–$223 daily API rate for the same workload. DeepSeek’s API pricing is $0.15 per 1M…
Key points
- Call Center Doctors rented 4x Nvidia H200s for $440.88/day to test DeepSeek V4.1 Flash, exceeding claimed 80x cost savings
- DeepSeek’s API costs $3,500–$7,000 for 388.5B tokens (vs. $5,500 for Claude’s flat-rate subscription)
- 96% of tokens were re-reads, and sandbox issues forced agents offline, making GPU rentals impractical
The test revealed inefficiencies: 96% of tokens were stale re-reads, and the agents struggled with sandbox escapes, forcing them offline. The consultancy’s September Claude bill was $5,500, while DeepSeek’s API would cost $3,500–$7,000 for the same tokens. Even accounting for DeepSeek’s lower per-token cost, the consultancy concluded renting GPUs was unfeasible. DeepSeek V4.1-Pro, which may improve performance, has no release date yet.
Model pages: DeepSeek-V4.1-Flash → · Claude Opus 5.5 →
Firm rents four Nvidia H200s to test '80x cheaper' DeepSeek claim
Tom's Hardware · 1 October 2026
Loading the full article…
This text was published by Tom's Hardware and written by Shane Downing. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Synopsys unveils Autopilot platform for AI-driven chip design agents · 3 src
- NVIDIA’s NVFP4 cuts prompt processing time by 46% on Blackwell GPUs · 1 src
- Alphabet to launch Google AI chips into orbit on SpaceX Falcon 9 · 1 src
- Shure launches IntelliMix Bar Pro with Microsoft Teams certification · 1 src
- ASML’s EUV monopoly powers Nvidia’s AI chip demand · 2 src
Comments
via GitHub Discussions