Ask
Price change

Novita cuts Macaron V1 Venti and Tall token rates 45%

Novita AI pricing

Mind Lab's Macaron V1 Venti and Macaron V1 Tall both dropped 45% on Novita's serverless rate card, less than a month after converting from a free launch tag to paid per-token pricing.

Before

Macaron V1 Venti $1.5/M in ($0.3/M cache read) · $4.5/M out; Macaron V1 Tall $0.45/M in ($0.08/M cache read) · $2.6/M out (rates set when both models converted off their 'Time Limited Free' tag on 2026-08-11, still in effect as of the 2026-08-26 capture).

After

Macaron V1 Venti $0.825/M in ($0.165/M cache read) · $2.475/M out; Macaron V1 Tall $0.2475/M in ($0.044/M cache read) · $1.43/M out — every field on both models scaled to exactly 55% of its prior value.

Both Macaron models moved by the identical 55% multiplier across input, output, and cache-read rates alike — input, output, and cache-read all landing at exactly 55% of their 2026-08-11 conversion price — which points to a deliberate blanket repricing of the Mind Lab family rather than a per-field adjustment. It is the fastest reprice-after-launch cycle tracked for these two SKUs: converted from free to paid on 2026-08-11, then cut 45% again less than four weeks later, on 2026-09-07.

From Novita AI's pricing timeline
RTX 5090 cut 23%; H100 pulled from self-serve again; Image Endpoints and sandbox credit removed

Just 12 days after H100 returned to self-serve, the /en/gpus page removed it again — H100 SXM 80GB stays available only via the $1.99/GPU-hr dedicated endpoint and $1.70/GPU/hr bare-metal routes. The self-serve RTX 5090 32GB base tier was cut ~23% to $0.56/hr (no spot rate shown), while a separate 'High frequency' RTX 5090 tier was reinstated at $0.72/hr on-demand / $0.36/hr spot. The same pass dropped Dedicated Endpoints' Image Endpoint subscriptions ($559/mo Standard, $1,199/mo Pro) and Agent Sandbox's $100/90-day signup-credit offer.

About Novita AI
novita.ai ↗

Novita AI is a pay-as-you-go AI cloud offering inference across 174 listed models as of the 2026-08-14 capture (down from 177 at 2026-08-04/2026-08-11, after that 2026-08-04 refresh added Deepseek V3.2, Deepseek V4 Flash, DeepSeek-OCR 2, and PaddleOCR-VL, following a 2026-07-29 delisting of legacy video/image SKUs including Hunyuan Video Fast, PixVerse V4.5, and the Kling-o1 lineup), on-demand and bare-metal GPUs, and secure per-second agent sandboxes under a single API; a 2026-08-25 capture cut the Image catalog from 14 to 5 SKUs and pruned most legacy Video lines while the site's own model-catalog counter read 144, a divergence from the 174 figure this page has tracked that remains unreconciled.

Free tier
Yes
Commits
None
Transparency
public

Novita AI pricing history

  1. Sep 2026
    RTX 5090 cut 23%; H100 pulled from self-serve again; Image Endpoints and sandbox credit removed
  2. Sep 2026
    Macaron V1 Venti and Tall cut 45% across input, output, and cache-read
  3. Aug 2026
    H100 SXM and L40S return to self-serve GPU instances; RTX 6000 Ada and RTX 5090 HF variant removed
  4. Aug 2026
    Deepseek V4 Flash 0731 repriced sharply; RTX 6000 Ada returns; Image/Video catalog cut hard
  5. Aug 2026
    Ling 3.0 Tiny delisted after 3 days, emptying the free-LLM shelf
Full Novita AI timeline

More Novita AI activity

All pricing activity