DeepSeek's Vague Price Hike Notice Signals End of AI Inference's Race-to-the-Bottom Era

Discuss AI News
1
  • DeepSeek announced August 6 it plans to raise API prices ‘by a relatively large margin’ — no number, date, or explanation given
  • Media attribution to GPU cost pressure is speculation; the vague announcement format is itself a strategic signal
  • Analysis identifies multiple simultaneous motives: compute costs, user filtering, pre-funding valuation, and value-based pricing shift
  • Multiple LLM vendors have recently raised prices, cut free tiers, or suspended sign-ups — signaling a structural industry-wide shift

:link: Read the full story on Goblin News

2

This is crazy… nowhere is safe! I’ve moved a bunch of inference for my high volume functions to deep seek because it’s insanely cheap. “Significant” price increase out of nowhere was not something I expected. We’ll see how this goes. Maybe a significant increase on something that’s damn near free won’t actually even register.

image
image2336×478 51.7 KB

3

The new rates are out. Significantly higher for sure - but still quite lower than the major western labs across most offerings.

It’s either no big deal or a very big deal depending on how you frame it. The big deal scenario is if you’ve built user-facing features with DeepSeek models behind them. Your inference cost goes up ~2-5x instantly (depending on your volume at peak vs. off peak). And likely, you wouldn’t have accounted for that risk because API rates don’t normally go up within each generation of models. Getting the rug pulled on subscription subsidies is normal, but API rates have typically only gone down for existing model offerings.

This hits home especially because I have been building an app where the deepseek rates make certain features viable in a free tier. But with the rates shooting up (plus the instability risk that it highlights), I may have to rethink the approach.

I guess the silver lining is the lesson learned and the fun rabbit hole it will send me down to explore my options.

Model Token Type Old Flat Rate New Off-Peak New Peak Off-Peak vs Old Peak vs Old
V4-Flash Cache Hit (Input) $0.0028 $0.007 $0.014 2.5× 5.0×
Cache Miss (Input) $0.14 $0.22 $0.44 1.6× 3.1×
Output $0.28 $0.66 $1.32 2.4× 4.7×
V4-Pro Cache Hit (Input) $0.003625 $0.022 $0.044 6.1× 12.1×
Cache Miss (Input) $0.435 $0.66 $1.32 1.5× 3.0×
Output $0.87 $1.98 $3.96 2.3× 4.6×

Notes:

  • Peak hours: 01:00–04:00 & 06:00–10:00 UTC, Monday–Friday only (≈ Beijing 9:00–12:00 and 14:00–18:00)

  • Off-peak = all other times (including weekends) and is exactly half of peak

  • New rates took effect August 16, 2026

  • Even the cheaper off-peak rates are higher than the old flat rates in every category