BroadRouter

LLM API price changes in October 2026: 3 rises, no cuts

By the BroadNet Inference team · 8 min read · Published

Few, and no cuts. Comparing archived pricing pages of 7 LLM API providers from early October to early November 2026, we found 3 list-price rises: two at DeepInfra (Mistral Nemo, Gemma 4 31B) and one at Together AI, where Qwen3.8 Flash went back to its September price. 9 models were newly priced and 11 left the pages.

This is our recap of what changed on LLM API providers’ own pricing pages in October 2026. It compares Internet Archive captures of each page from early October with those from early November, model by model, and lists every price rise, cut, new price and removal we found, with links to both captures. BroadRouter is BroadNet’s website. BroadNet Inference is not in this data, because the archive has no captures of its pricing page.

The rows behind it are in our LLM API price history dataset. For the longer view, see how often LLM API prices change, LLM API price cuts 2023 to 2026 and do LLM API prices only go down?.

What period does this cover?

The window is each provider’s last two monthly captures: one taken on 1 to 3 October 2026 and one on 1 to 2 November 2026. A change listed here happened somewhere between those two dates. We call it October because that is the month the window spans; it says nothing about when a provider announced the change. November as a whole cannot be covered yet: the newest captures are from its first two days.

Seven of the 11 providers in the dataset have a usable capture at both ends:

Provider Earlier capture Later capture
OpenAI 1 Oct 2026 1 Nov 2026
Anthropic 2 Oct 2026 1 Nov 2026
Google (Gemini API) 1 Oct 2026 1 Nov 2026
DeepSeek 3 Oct 2026 1 Nov 2026
xAI 2 Oct 2026 1 Nov 2026
Together AI 3 Oct 2026 2 Nov 2026
DeepInfra 1 Oct 2026 2 Nov 2026

The other four are not covered this month. Mistral AI’s and Cohere’s archived pages no longer show API token prices in their HTML, Fireworks AI’s no longer lists prices per named model in a form we read, and the dataset has no Groq capture after 2 August 2026.

Which LLM API prices went up in October 2026?

Three, all at hosts of open-weight models. USD per 1M tokens, input / output. “Blended” is 3 input tokens per output token, as on the rest of BroadRouter.

Provider and model Before, in / out After, in / out Change Archive
DeepInfra Mistral-Nemo-Instruct-2407 0.019 / 0.03 0.029 / 0.03 input +52.6%, blended +34.5% 1 Oct, 2 Nov
DeepInfra gemma-4-31B-it 0.15 / 0.40 0.20 / 0.40 input +33.3%, blended +17.6% 1 Oct, 2 Nov
Together AI Qwen3.8 Flash 0.09 / 0.28 0.15 / 0.47 input +66.7%, output +67.9%, blended +67.3% 3 Oct, 2 Nov

Each has a back story in earlier captures:

  • gemma-4-31B-it at DeepInfra rose for the second month running. It was 0.13 / 0.38 in every capture from June to September 2026, 0.15 / 0.40 on 1 October and 0.20 / 0.40 on 2 November: input up 54% in two months.
  • Mistral-Nemo-Instruct-2407 at DeepInfra was 0.02 / 0.04 until July 2026, was cut to 0.019 / 0.03 in the August capture, and now has a higher input price than before that cut.
  • Qwen3.8 Flash at Together AI was 0.15 / 0.47 on 2 September, 0.09 / 0.28 on 3 October and 0.15 / 0.47 again on 2 November. The October capture did not mark the lower price as promotional, so the dataset counts both moves as list-price changes. It may have been a short discount that the page did not label; the captures cannot tell.

Were there any price cuts?

No. No model listed in both captures got cheaper on input or output at any of the 7 providers. OpenAI, Anthropic, Google, DeepSeek and xAI showed the same input and output prices for every model they listed in both captures.

Two rows changed without an input or output price change, so neither counts as a change. At DeepInfra, Qwen3.6-35B-A3B now shows a cached-input price of 0.10, equal to its input price of 0.10, where the October capture showed none; input and output stayed at 0.10 / 0.95. At Anthropic, Claude Sonnet 5.5 stayed at 2.00 / 10.00, but its cache-hit price went from 0.20 to 0.10.

Which models got new prices?

Nine models appear in a November capture with a price and were not on the same provider’s page in October:

Provider Model Price, in / out Cached input Note
OpenAI gpt-6.1-terra 1.00 / 5.00 0.05 Standard short-context price
OpenAI gpt-6-rosalind 15.00 / 75.00 1.50 Short-context price; life sciences model, access limited to approved research through a trusted-access program
OpenAI gpt-rosalind-discovery 5.00 / 25.00 0.50 Life sciences model, same access limit
Anthropic Claude Haiku 5.5 0.10 / 0.50 0.01 Price for prompts up to 100,000 tokens; the page lists 0.50 / 2.50 for longer prompts
Google (Gemini API) Gemini 4 Argon 2.00 / 10.00 0.10 Promotional: “through November 30, 2026”; the page lists 4.00 / 20.00 “starting December 1, 2026”
Together AI MiniMax M3.1 Flash 0.15 / 0.60 0.015
Together AI DeepSeek V4.1 Flash 1017 0.30 / 1.20 0.006
Together AI NVIDIA Nemotron 3.5 Super VL 0.15 / 0.50 0.02
DeepInfra NVIDIA-Nemotron-3.5-Super-VL 0.15 / 0.50 0.02

All nine come from the November captures linked in the period table above. Gemini 4 Argon is the one to watch: if Google keeps the rate its page states, the list price doubles on 1 December 2026. NVIDIA Nemotron 3.5 Super VL arrived at both hosts at the same price. For how one open model is priced across hosts, see same LLM, different host prices.

Which models were removed from pricing pages?

Eleven models priced in an early-October capture were missing from the same provider’s page in early November. In each case our check of the live page on 14 November 2026 did not list the model either. Leaving a pricing page does not by itself mean the API stopped serving the model; it means the provider no longer lists a price for it there.

Provider Model Last listed price, in / out (October capture)
Google (Gemini API) Gemini 3.7 Flash 0.75 / 3.75 (marked promotional, “through December 31, 2026”)
Google (Gemini API) Gemini 3.5 Flash 1.50 / 9.00
Together AI Gemma 4 31B 0.39 / 0.97
Together AI DeepSeek V4 Flash 0731 0.14 / 0.28
Together AI Llama 3.3 70B 1.04 / 1.04
Together AI MiniMax M2.7 0.30 / 1.20
DeepInfra Qwen3-Max-Thinking 1.20 / 6.00
DeepInfra Qwen3-Max 1.20 / 6.00
DeepInfra Meta-Llama-3.1-70B-Instruct-Turbo 0.40 / 0.40
DeepInfra Nemotron-3-Nano-30B-A3B 0.05 / 0.20
DeepInfra MythoMax-L2-13b 0.40 / 0.40

Before and after captures: Google 1 Oct and 1 Nov; Together AI 3 Oct and 2 Nov; DeepInfra 1 Oct and 2 Nov. On Google’s November page, Gemini 3.7 Flash and 3.5 Flash still appear as links to their model pages, but not in the price tables.

MythoMax-L2-13b had been priced in every DeepInfra capture in the dataset since the first, in April 2024.

What changed the month before?

For context, the previous window (captures of early September to early October 2026) had 3 cuts and 3 rises. DeepSeek’s September capture was not usable, so its window runs from 1 August.

Provider and model Captures Before, in / out After, in / out Blended change Archive
DeepSeek deepseek-v4-pro 1 Aug → 3 Oct 2026 0.435 / 0.87 1.32 / 3.96 +264.1% before, after
Together AI Qwen3.7-Max 2 Sep → 3 Oct 2026 1.25 / 3.75 1.50 / 4.50 +20.0% before, after
DeepInfra gemma-4-31B-it 1 Sep → 1 Oct 2026 0.13 / 0.38 0.15 / 0.40 +10.4% before, after
Together AI Qwen3.8 Flash 2 Sep → 3 Oct 2026 0.15 / 0.47 0.09 / 0.28 −40.2% before, after
DeepInfra NVIDIA-Nemotron-3.5-Lightning 1 Sep → 1 Oct 2026 0.08 / 0.20 0.06 / 0.16 −22.7% before, after
DeepInfra DeepSeek-V4-Flash-0731 1 Sep → 1 Oct 2026 0.08 / 0.18 0.06 / 0.18 −14.3% before, after

The pattern matches the long-run data: model makers rarely reprice a model they already list, while hosts move prices in both directions, sometimes month after month.

What are the limits of this recap?

  • List prices only. These are the standard pay-as-you-go text prices on each provider’s public page. Batch, flex and priority tiers, volume discounts, credits and negotiated contracts are not included, so they are not what any given customer pays.
  • Monthly snapshots. A change is only known to fall between two captures. A price that changed and changed back between captures would not show.
  • Coverage. 7 of the 11 providers in the dataset had usable captures at both ends of the window. Nothing here describes the other four, or any provider outside the dataset.
  • No usage data. We count changes on pricing pages. Nothing here says how many people use a model.

Every row, with its archive link and the exact text it was read from, is on the dataset page and in the CSV. Live changes from our own price checks are in the LLM API price change log.

Where BroadNet Inference fits

BroadRouter is BroadNet’s own site. BroadNet Inference is not in the archived-price data, because the Internet Archive had no captures of its pricing pages when we checked on 15 November 2026, so this recap cannot say whether its prices moved. Its current rates are published in a machine-readable pricing.json. For a fixed monthly cost instead of a per-token rate, BroadNet also sells a flat plan for DeepSeek V4.1 Flash at $60 a month per session (one request at a time); see flat rate vs pay per token.

Questions

Did any LLM API prices go down in October 2026?

Not in our captures. Comparing the early-October and early-November 2026 captures of 7 providers' pricing pages, we found no list-price cut on a model that stayed listed. The month before (captures of September to early October) had 3 cuts and 3 rises.

Which LLM API prices went up in October 2026?

Three, all at hosts. DeepInfra raised Mistral-Nemo-Instruct-2407 input from USD 0.019 to USD 0.029 per 1M tokens and gemma-4-31B-it input from USD 0.15 to USD 0.20. Together AI raised Qwen3.8 Flash from USD 0.09 / USD 0.28 to USD 0.15 / USD 0.47, the price it showed in September.

Is Gemini 4 Argon's price going up?

Google's pricing page, as captured on 1 November 2026, lists Gemini 4 Argon at USD 2.00 input and USD 10.00 output per 1M tokens "through November 30, 2026" and USD 4.00 and USD 20.00 "starting December 1, 2026". We record the current price as promotional; the December rate is the page's own statement.

Why is this a recap of October and not November 2026?

Our data comes from monthly Internet Archive captures. The latest ones are from 1 to 2 November 2026, so the newest complete window runs from the early-October captures to those. Changes made later in November will show in the next captures.

Is BroadNet Inference in this data?

No. BroadRouter is BroadNet's website, and the Internet Archive had no captures of BroadNet's pricing pages when we checked on 15 November 2026, so BroadNet has no archived price history to compare.

Read as MarkdownAsk ChatGPTAsk ClaudeAsk Perplexity