ModelPriceWatch.com
Last scan 2026-08-15 Models tracked 205 Providers 31 Cheapest paid Granite 4.0 H Micro $0.017/Mtok in Every price links to its source

IBM price history

Every pricing change we record for IBM API models, newest first. Tracking began Oct 20, 2025. Prices re-verified Aug 15, 2026.

What changed

3 recorded changes · newest first · every move keeps its receipt
  1. Aug 14, 2026

    Data correction: Granite 4 H Small data correction

    Published $0.06/$0.25 was never IBM's printed price: it traces to a 2025-10-20 litellm-archive backfill (confidence: reconstructed) and was never read off IBM. IBM's watsonx.ai rate card — this row's own pricing_url, and its own committed 2026-08-09 receipt — prints granite-4h-small at USD 0.0636 per 1M tokens input / USD 0.265 output, and printed the same on us-en on 2026-08-14. The 6% gap is IBM's own uniform uplift over round base rates, applied across the whole table (granite-8b-code-instruct USD 0.636, granite-guardian-3-8b USD 0.212, llama-3-3-70b USD 0.7526, mistral-small-3-1 USD 0.106/0.318) — not tax (the footnote excludes taxes and duties) and not country variance (us-en is identical). The sibling row ibm-granite-embedding-278m already publishes the uplifted USD 0.106 off this same table. IBM did not re-price: this is our number being corrected to the rate card, not a market move. (input_per_mtok: $0.06 → $0.0636 · output_per_mtok: $0.25 → $0.265)

  2. Aug 14, 2026

    Model removed: Granite 4 H Large removed

    Removed because no such model exists. IBM's watsonx.ai rate card lists every Granite SKU it sells — including the ones with no per-token price — and there is no Granite 4 H Large on it; IBM never announced one, and no copy of the weights exists on Hugging Face. The $0.30/$1.20 we carried was a seed default from our first bulk import, not a price anyone charges.

  3. Aug 14, 2026

    Model removed: Granite 4 H Medium removed

    Removed because no such model shipped. IBM named Granite 4.0 Medium as a planned future model at the Granite 4.0 launch and never released it — it is absent from IBM's rate card and from Hugging Face, and IBM's next release was the Granite 4.1 line instead. The $0.15/$0.60 we carried was a seed default from our first bulk import, not a price anyone charges.

Current IBM prices

3 models · cheapest blended first

Per 1M tokens, cheapest blended first. "First tracked" is when this model entered our price history.

Current IBM model prices per 1M tokens with the date each entered our price history
Model Blended* Input Output First tracked
Granite 4.0 H Micro $0.041 $0.017 $0.112 Jun 25, 2026
Granite Embedding 278M Multilingual $0.106 $0.106 $— Jun 25, 2026
Granite 4 H Small $0.114 $0.064 $0.265 Oct 20, 2025
* Blended = (3×input + 1×output) ÷ 4 $/1M tokens MODELPRICEWATCH.COM · 2026-08-15

About this log

ModelPriceWatch re-verifies IBM API pricing against its official pricing page on a twice-daily refresh cycle and records every change here.

Quick stats

Models tracked
3
Recorded changes
3
Tracking since
Oct 20, 2025
Last change
Aug 14, 2026