Updated weeklypricing

AI Pricing This Week: Large 4 Cut, New Hosts, Oct 26-Nov 1

Short answer. From October 26 to November 1, 2026, one list price moved: Mistral cut Mistral Large 4 to $0.70 / $2.10 per million tokens, first seen on October 30. Two hosts of open models, Novita AI and Baseten, joined the dataset. GPT-6.1 Terra arrived at $1.00 / $5.00. Read after the week: Mistral deprecated Medium 3.5 and Large 3.

In plain words. One model got cheaper, two new sellers of open models joined our table, and one OpenAI model arrived. Mistral marked two older models for removal, which we saw the following week. Prices here are per million tokens. A token is a short piece of text, about three-quarters of a word.

This is the third weekly summary. It covers Monday, October 26 to Sunday, November 1, 2026, plus what we read on November 2 and 3, which is marked as such. Every fact comes from AI Signal’s dataset and change log, with the provider page it was read on and the date.

The week in three lines. Mistral Large 4’s price fell by about half on October 30, and the same page added a DeepSeek model served by Mistral. Novita AI and Baseten joined the dataset as hosts on October 29, and with them the widest price gap we track, 7.3× on DeepSeek V4 Flash 0731. OpenAI released GPT-6.1 Terra on October 26, at half of GPT-6.1 Sol’s price.

Shutdowns, read after the week closed

GLM 5.2 on Mistral, October 31. Mistral’s changelog says: “Z.ai GLM 5.2 ( zai-glm-5-2 ) is deprecated and retires on October 31, 2026. Use Z.ai GLM 5.3 ( zai-glm-5-3 ) instead, at the same price.” We read that line on November 3, 2026, and marked GLM 5.2 as retired that day, with the price it had; GLM 5.3 is on Mistral’s pricing page at the same $1.40 / $4.40.

Mistral Medium 3.5 and Mistral Large 3, deprecated on October 31. Our weekly check of November 2 was the first to find neither model on Mistral’s pricing page. On November 3 their model pages said “This model is deprecated. Use Mistral Large 4 for new integrations.”, “Deprecation date 10/31/2026” and “Replacement Mistral Large 4”, with no retirement date on the page as shown in a browser. Both rows are marked deprecated, with the last prices read on October 26: $1.50 / $7.50 for Medium 3.5 and $0.50 / $1.50 for Large 3. What moving to Large 4 costs from each is in Mistral deprecates Medium 3.5 and Large 3.

Price changes

Mistral Large 4, October 30. The only list price that moved. The weekly check of October 26 read Mistral’s pricing page as “Mistral Large 4 ↗ | $1.36 | $0.14 | $4.18 |”; on October 30 the same table printed “Mistral Large 4 ↗ | $0.7 | $0.07 | $2.1 |”: $0.70 per million input tokens, $0.07 for cached input and $2.10 for output, 1.36 ÷ 0.70 = 1.94× and 4.18 ÷ 2.10 = 1.99× lower. Mistral does not say when it changed the price or whether the new one is temporary: its changelog still prints “Launch pricing: 50% off for 2 weeks.” under October 6, and nothing later. The price-change log dates it October 30, the first day it was seen; the post is Mistral Large 4’s price cut.

Kimi K3’s cached input on Together AI, recorded October 29. Together raised the cached-input price of Kimi K3 from $0.27 to $0.30 per million tokens; input ($3.00) and output ($15.00) did not move. The last reading at $0.27 was a hand reading of October 12; the next saved reading, October 17, already printed “Kimi K3 | $3.00 $0.30 (cached) | $15.00 |”, as did the pages of October 19, 26 and 29. The entry was written on October 29 and dated October 17, the first day the new figure was seen.

Our own corrections

None of these was a price change by a provider; each entry in the log says so in its first line.

  • OpenAI’s “All models” table, October 26. The folded table that our reader had not seen, described in last week’s summary. From it, GPT-5, GPT-5 mini and o3 got their prices back and fourteen rows OpenAI already priced entered the dataset that Monday (see New models).
  • Cached input on two hosts, October 29. DeepInfra’s pricing page prints a cached-input price inside the input cell, and Fireworks AI’s prints each price as input, cached input and output; four of our offerings did not carry the figure the page printed. DeepSeek V4 Flash 0731 ($0.015) and Kimi K3 ($0.285) on DeepInfra, gpt-oss-120b ($0.015) and MiniMax M3 ($0.06) on Fireworks AI now do. Input and output prices did not change.
  • GPT-5.6 Cyber’s cache writes, October 29. The Markdown version of OpenAI’s pricing page prints ”| gpt-5.6-cyber | $12.50 | $75.00 | $1.25 | $15.625 | $31.25 | - | - | - | - | - |”: a 30-minute cache write of $15.625 and a 12-hour one of $31.25 per million tokens, which GPT-5.6 Cyber’s row did not carry.

Since October 30 the weekly price check also compares the cached-input price of each row with the quote it reads from the page, where the page prints one; a quote that marks a different cache price is reported, not stamped.

New models

ModelReleasedAdded hereInput / output per 1M tokens
GPT-6.1 TerraOctober 26October 27$1.00 / $5.00
DeepSeek V4.1 Flash 1017 on MistralOctober 28October 30$0.30 / $1.20

OpenAI’s pricing page, read on October 27, prints ”| gpt-6.1-terra | $1.00 | $5.00 | $0.05 | $1.25 | $2.50 | $2.00 | $7.50 | $0.10 | $2.50 | $5.00 |”: input, output and cached input, the two cache writes, then the same five for prompts over 272,000 tokens. Every one of those prices is half of GPT-6.1 Sol’s. The post is GPT-6.1 Terra API pricing.

Mistral’s pricing page, read on October 30, prints “DeepSeek V4.1 Flash 1017 ↗ | $0.3 | $0.007 | $1.2 |”, and the model’s page says “October 28, 2026 License Public Preview Third-party MIT v v4.1-flash-1017” and “The model is served without Mistral modifications.” It is a DeepSeek model sold by Mistral at one price, without DeepSeek’s own peak and off-peak tariff, and it is listed apart from Mistral’s own models on the Mistral API pricing page.

On Monday, October 26, fourteen OpenAI rows entered from the “All models” table, as told last week: five current (GPT-5.4 mini, GPT-5.2, GPT-5.2 Pro, GPT-5.5 Cyber and GPT-5 Search API) and nine deprecated. They are not new at OpenAI; they were missing here.

New hosts

On October 29 two hosts of open models entered the dataset, each read on its own pricing page: Novita AI (novita.ai/pricing), for the models that already have a page on this site, and Baseten (baseten.co/pricing). Novita AI’s page prints, for example, “DeepSeek V4 Flash 0731 | 1M | $0.44 /Mt · Cache Read $0.028 /Mt | $1.32 /Mt | More |”; Baseten’s prints each name and figure twice, “DeepSeek-V4-Flash-0731 $0.13 $0.13 $0.028 $0.028 $0.26”.

With them, three open models cost more than 1.15× as much on one host as on another, read on October 29, 2026:

Open modelLowest input priceHighest input priceHighest ÷ lowest
DeepSeek V4 Flash 0731$0.06, DeepInfra$0.44, Novita AI7.3×
gpt-oss-120b$0.05, Novita AI$0.15, Fireworks AI, Groq and Together AI3.0×
Llama 3.3 70B Instruct$0.10, DeepInfra$0.135, Novita AI1.4×

The same reading is the subject of the same open model at a different host price. Novita AI serves Llama 3.3 70B with a 12,288-token context, where DeepInfra serves 131,072; the inference providers page says so under the row.

Subscriptions and coding tools

No price moved. On October 27 the pages read by hand from Spain were read again. ChatGPT’s plans window, which on October 17 printed Pro as “Desde €103 / mes” and “Disponible en 3 niveles de uso”, now shows the three levels, €103, €229 and €510 a month, VAT included; the post is ChatGPT’s price in Spain, every plan. Grok’s plans page no longer shows the first-month offer: Premium is “31 €/mes”, with no discount, the full price this site already published. Claude’s and Google’s Spanish pages, Perplexity’s and the Claude Code page were read the same day with no change in price.

New pages

Three maker pages went up on Monday, October 26: Mistral, xAI and Meta, each with the maker’s current models, cheapest first, and the models it serves for others listed apart.

Other posts of the week: Mistral Pro (Le Chat) vs ChatGPT Plus (October 26) and what open-model hosts say besides the token price (October 28): credits, minimums and sales contacts on Together, Fireworks, DeepInfra, Groq and Novita.

After the week closed: Claude’s new Ultra plan and the end of Max 20x, and Google’s shutdown dates for Gemini 2.5 Pro and 2.5 Flash, were seen between November 3 and 6 and belong to the next summary; the posts are Claude Ultra and Gemini 2.5 Pro and 2.5 Flash shut down on December 1.

What comes next

Dates as the providers print them, all on Deprecations:

  • November 21: the last day OpenAI promises GPT-5.6 Sol’s price: “GPT-5.6 Sol’s promotional pricing is available at least through November 21, 2026.” Update, November 9, 2026: OpenAI’s pricing page and GPT-5.6 Sol’s model page, read on November 9, 2026, now say “GPT-5.6 Sol’s promotional pricing is available until November 30, 2026.” Up to the weekly reading of November 2 the wording was “at least through November 21, 2026”. The price, $4.00 and $20.00, is unchanged, and OpenAI still publishes no list price for after it; the change is recorded on Price changes.
  • November 30: Anthropic retires Claude Sonnet 4.5: “November 30, 2026 | claude-sonnet-4-5-20250929 | claude-sonnet-5-5 |”. The same day is the last of Gemini 4 Argon’s introductory price: “$2.00 through November 30, 2026. $4.00 starting December 1, 2026.”
  • December 11: OpenAI shuts down GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 Pro, o3 and o3-pro.

How we checked

The facts were listed by our weekly script from AI Signal’s own data files: the models dataset, the price-change log, the deprecation records, the subscription and coding-tool data and the posts of the week. Every price and date was read on the provider’s own page, on the date given, November 2 and 3 included. Model prices are list prices in US dollars per million tokens. We do not know why any of these prices are what they are, and this summary does not guess.

Frequently asked questions

Which AI model prices changed between October 26 and November 1, 2026?

One list price: Mistral cut Mistral Large 4 from $1.36 to $0.70 per million input tokens and from $4.18 to $2.10 for output, first seen on October 30, 2026. Together AI's cached-input price for Kimi K3, $0.30 instead of $0.27, was recorded on October 29 with the first reading that showed it, October 17.

Which models were deprecated or shut down that week?

Z.ai's GLM 5.2 on Mistral retired on October 31, 2026, and Mistral deprecated Mistral Medium 3.5 and Mistral Large 3 on the same date, pointing both to Mistral Large 4. AI Signal read the three on November 2 and 3, after the week closed.

Which new models and hosts were added that week?

GPT-6.1 Terra at $1.00 / $5.00 per million tokens on October 27, DeepSeek V4.1 Flash 1017 on Mistral at $0.30 / $1.20 on October 30, and fourteen OpenAI rows from its All models table on October 26. Two hosts of open models entered the dataset on October 29, Novita AI and Baseten.

Written with AI assistance from this site's own verified pricing dataset and reviewed by Álvaro Lucero Rufino, who is responsible for what stays published here. Every figure links to the page where it was verified, with the date it was read; if a number here and a number on that page ever disagree, the page is the one to trust. Automated publishing was discontinued in September 2026; posts that no longer met that standard were removed rather than left up.