Tokens & Costs · Updated daily
AI pricing changes: the daily LLM API price log
Every change to AI API prices in our data, newest first: price cuts, increases, new models and removed ones, with the old and new price per million tokens. Data clean-ups and routing noise are filtered out, and list prices are kept apart from open-weight host prices. Last checked 11 October 2026.
Tracking since 8 October 2026Last change 11 October 2026Atom feed
Summary
Biggest moves in October 2026
Net change per model this month, combining every logged change, measured on the blended price (three parts input to one part output, per million tokens). Tracking began on 8 October 2026.
List prices
Closed-weight models, priced by their developer.
No closed-weight model’s list price changed in October 2026 so far. None has changed since tracking began on 8 October 2026.
Open-weight host prices
The price of OpenRouter’s top-ranked host, which moves when hosts change.
Cuts
- DeepSeek V4 Pro 0423$1.1854 → $0.261 blended−78%
- Qwen3 VL 30B A3B Thinking$0.75 → $0.4675 blended−38%
- Qwen3.5-35B-A3B$0.3625 → $0.2475 blended−32%
Increases
- Llama 3.3 70B Instruct$0.155 → $0.29 blended+87%
- Qwen3 VL 8B Instruct$0.2015 → $0.375 blended+86%
- Qwen3 30B A3B Instruct 2507$0.0844 → $0.15 blended+78%
Log
Every change, by day
46 changes over 4 days.
3 price cuts, 2 increases, 3 new models, 1 removed
- Input
- from $0.9483 to $0.2088
- −78%
- Output
- from $1.8966 to $0.4176
- −78%
- Cached input
- from $0.079 to $0.0174
- −78%
- Input
- from $0.465 to $0.65
- +40%
- Output
- from $2.45 to $3.41
- +39%
- Cached input
- from $0.0875 to $0.15
- +71%
- Input
- from $0.64 to $0.80
- +25%
- Output
- from $13.50 to $10
- −26%
- Cached input
- from $0.28 to $0.55
- +96%
- Input
- from $0.15 to $0.10
- −33%
- Output
- from $1 to $0.95
- −5.0%
- Cached input
- from $0.05 to $0.10
- +100%
- Input
- from $0.50 to $0.60
- +20%
- Output
- from $2.20 to $2.40
- +9.1%
- Cached input
- from $0.10 to $0.12
- +20%
New model
DeepSeek V4 Pro 0813DeepSeek
First listed by our source on this day.
New model
Hy3Tencent
First listed by our source on this day.
New model
Hy4 previewTencent
First listed by our source on this day.
Removed
qwen3.8-27bQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
3 price cuts, 8 increases, 11 removed
- Input
- from $0.117 to $0.25
- +114%
- Output
- from $0.455 to $0.75
- +65%
Price increaseOpen weights · host price
Qwen3 235B A22B Thinking 2507Qwen
Blended price $0.7475 → $1.2125 +62%
- Input
- from $0.23 to $0.45
- +96%
- Output
- from $2.30 to $3.50
- +52%
Price increaseOpen weights · host price
Nemotron 3.5 LightningNVIDIA
Blended price $0.0687 → $0.1025 +49%
- Input
- from $0.0469 to $0.07
- +49%
- Output
- from $0.134 to $0.20
- +49%
- Cached input
- from $0.0235 to $0.04
- +71%
- Input
- from $0.20 to $0.29
- +45%
- Output
- from $2.40 to $1
- −58%
- Output
- from $0.42 to $0.80
- +90%
- Cached input
- from $0.135 to $0.14
- +3.7%
- Input
- from $0.09 to $0.0675
- −25%
- Output
- from $0.30 to $0.225
- −25%
- Cached input
- from $0.05 to $0.0375
- −25%
- Input
- from $0.43 to $0.50
- +16%
- Output
- from $1.75 to $2
- +14%
- Cached input
- from $0.08 to $0.10
- +25%
- Input
- from $0.50 to $0.64
- +28%
- Output
- from $12 to $13.50
- +13%
- Cached input
- from $0.30 to $0.28
- −6.7%
- Input
- from $0.45 to $0.50
- +11%
- Output
- from $2.25 to $2.50
- +11%
- Cached input
- from $0.07 to $0.15
- +114%
Price cutOpen weights · host price
Kimi K2.6Moonshot AI
- Cached input
- from $0.0975 to $0.0875
- −10%
- Input
- from $0.45 to $0.32
- −29%
- Output
- from $2.70 to $3.20
- +19%
Removed
qwen-plus-2025-07-28Qwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-235b-a22bQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-30b-a3b-thinking-2507Qwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-8bQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-coder-plusQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-maxQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-max-thinkingQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-vl-235b-a22b-thinkingQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-vl-32b-instructQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3-vl-8b-thinkingQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
Removed
qwen3.6-max-previewQwen
No longer listed by our source. This doesn’t always mean the provider retired it.
4 price cuts, 7 increases, 2 new models, 1 removed
- Input
- from $0.10 to $0.22
- +120%
- Output
- from $0.32 to $0.50
- +56%
Price increaseOpen weights · host price
Qwen3 30B A3B Instruct 2507Qwen
Blended price $0.0844 → $0.15 +78%
- Input
- from $0.0482 to $0.10
- +108%
- Output
- from $0.1931 to $0.30
- +55%
- Input
- from $0.30 to $0.45
- +50%
- Output
- from $2 to $2.70
- +35%
- Input
- from $0.019 to $0.029
- +53%
- Input
- from $0.15 to $0.08
- −47%
- Output
- from $1 to $0.75
- −25%
- Cached input
- from $0.05 to $0.04
- −20%
- Input
- from $0.15 to $0.10
- −33%
- Output
- from $1.50 to $1.10
- −27%
- Input
- from $0.45 to $0.55
- +22%
- Output
- from $3 to $3.50
- +17%
- Cached input
- from $0.22 to $0.225
- +2.3%
- Input
- from $0.085 to $0.08
- −5.9%
- Output
- from $0.40 to $0.45
- +12%
- Input
- from $0.62 to $0.50
- −19%
- Output
- from $12.30 to $12
- −2.4%
- Cached input
- from $0.43 to $0.30
- −30%
- Input
- from $0.049 to $0.0469
- −4.3%
- Output
- from $0.14 to $0.134
- −4.3%
- Cached input
- from $0.0245 to $0.0235
- −4.3%
- Input
- from $0.2574 to $0.32
- +24%
- Output
- from $1.0287 to $0.89
- −13%
New model
Ling 3.0 Flash SanteinclusionAI
First listed by our source on this day.
New model
Step 5 PreviewStepFun
First listed by our source on this day.
Removed
ernie-4.5-vl-424b-a47bbaidu
No longer listed by our source. This doesn’t always mean the provider retired it.
1 price cut
- Input
- from $0.72 to $0.62
- −14%
- Output
- from $15 to $12.30
- −18%
Prices in US dollars per million tokens. Showing the last 90 days. Model names link to the model page, or to the comparison table when there is no page.
Method
How this log works
Where the data comes from
Once a day (scheduled for 05:17 UTC) a GitHub Action downloads the OpenRouter models API and the open-source LiteLLM price list, validates every record and compares it with the previous day. Each difference in a price (input, output, cached input, cache write or batch) and each model that appears or disappears is written to a change file. This page and its feed are rebuilt from that file, so every row is something the sources actually reported on that day.
List prices and host prices
OpenRouter’s documentation describes a model’s pricing field as the “pricing from the top provider for this model” (checked 11 October 2026). A closed-weight model such as GPT, Claude or Gemini is served only by its developer and the developer’s cloud partners, and OpenRouter states it adds no markup to inference, so a change there reflects the developer’s list price. Open-weight models are hosted by many companies at different prices, and the top-ranked host changes often. In our data, Kimi K3’s listed price moved on 4 of the 4 days logged (input $0.72 → $0.62 → $0.50 → $0.64 → $0.80 per million tokens). That is why open-weight rows are labelled host prices and summarised separately.
What we filter out
The raw change file holds 120 entries. 28 of them are hidden here because they are not price decisions:
- Our own corrections (4). When a source reports a wrong price, we fix it by hand from the provider’s pricing page (see the methodology). The fix shows up in the raw file as a change, but the provider’s price never moved.
- Prices that appear or disappear (13). A cached or batch price going from “not published” to a number, or back, usually means a source added a field or one of our sanity checks dropped an implausible value.
- Same-day reversals (8). When the data refreshes more than once in a day, the runs are combined, and a price that went down and back up again is not a change.
- Moves under 1% (3). In our data these were uniform shifts of every price of a model by the same fraction, typical of rounding or currency conversion.
How to read an entry
Each row gives the old and new price per million tokens and the change as (new − old) ÷ old. When input or output moved, the blended price weighs input three times as heavily as output, because typical chat and coding requests send several times more tokens than they get back. It decides whether a day counts as a cut or an increase when input and output move in opposite directions.
For example, on 11 October 2026 Kimi K3 went from $0.64 input and $13.50 output to $0.80 and $10. Blended: (3 × $0.64 + $13.50) ÷ 4 = $3.855 before, and $3.10 after, a change of −20%, so it is logged as a price cut.
New and removed models
“New model” means our source listed it for the first time. “Removed” means it left the source’s public list or failed a price check; it isn’t proof that the provider retired it. For shutdown dates, the model ID reference tracks the providers’ own deprecation pages.
Report an error
If a row looks wrong, email hello@its-tahir.com or use the contact page with the model, the date and the provider’s pricing page. Always confirm a price on the provider’s own pricing page before you rely on it.
FAQ
Frequently asked questions
How often do AI API prices change?
List prices set by a model’s developer change far less often than hosted prices. In the 4 days we have logged since 8 October 2026, we recorded 0 list-price changes for closed-weight models and 28 hosted-price changes across 22 open-weight models.
Why does an open-weight model’s price change so often?
Open-weight models such as Llama, Qwen or Kimi are served by many hosting companies at different prices. OpenRouter’s model list shows the price of the top-ranked host for each model, so the listed price moves whenever that ranking changes, even if no host changed its own price. We label these as host prices and keep them apart from list prices.
Does a price cut here lower my bill?
Only if you buy through the route that changed. If you call a developer’s API directly, you pay its list price, so check the provider’s pricing page. If you use OpenRouter, you pay the host your request is routed to. A cut to an open-weight model’s host price may apply to one host only. Re-run your numbers in the LLM cost calculator.
What does “removed” mean?
The model is no longer in our data: it left OpenRouter’s public model list, or failed one of our price checks. That doesn’t always mean the provider retired it. Sometimes a model is only relisted under a new or dated name. For official shutdown dates, see our model ID reference and the provider’s deprecation page.
How is the percentage change calculated?
Each row shows the old price, the new price and (new − old) ÷ old. When input and output move together we also show the blended price, three parts input to one part output, which is the typical mix for chat and coding work. Same-day refreshes are combined, and moves under 1% are hidden as noise.
Is there an RSS or Atom feed?
Yes. /pricing-changes/feed.xml is an Atom feed with one entry per day that had changes, listing each model with its old and new prices. Any feed reader can follow it. It is rebuilt with the site after each daily data refresh, so it never needs your email address or an account.
I think a change is wrong. How do I report it?
Email hello@its-tahir.com with the model, the date and a link to the provider’s pricing page, or use the contact page. If our source was wrong, we add a correction with the source link, and the change disappears from this log because it was a data fix, not a price move.
Related
Related tools
- LLM API Cost CalculatorEstimate per-request, daily and monthly API costs.
- AI Model ComparisonCompare prices, context windows and features across models.
- AI Model Pricing PagesSpecs, real costs and cheaper alternatives for popular models.
- Batch API Savings CalculatorCompare batch and real-time API costs.
- Prompt Caching CalculatorEstimate savings from prompt caching.
- Subscription vs API CalculatorFind out whether a chat plan or the API is cheaper for you.