Context Window · 18 October 2026 Model releases, pricing and API changes from OpenAI, Anthropic and other AI labs. Five updates, with source links below. Jugurtha-4B: A 4.2B-parameter model supports Kabyle conversation, translation and reasoning. https://huggingface.co/agbalu/Jugurtha-4B

Context Window: today's updates Qwen3.5-4B big-endian: Byte-swapped Qwen3.5-4B GGUF weights bring the model to IBM Z and LinuxONE. https://huggingface.co/taronaeo/Qwen3.5-4B-BE-GGUF xAI TypeScript SDK: Version 0.2.4 adds 21:9 and 5:2 video generation ratios and Grok Voice 3.0 model IDs.

DeepSeek added a price for Fetch, reading "Per 1,000 successful requests | $0.75".

OpenAI changed the retirement line for gpt-4o. It said "null" and it now says "2027-01-14". The same edit removed gpt-4o-mini ("null").

DeepSeek added a price for Search, reading "Per 1,000 successful requests | $0.75".

Google DeepMind changed the price it lists for Cache tokens. It said "$1.50 | $1.50" and it now says "$0.15 | $0.15".

xAI removed Speech to Speech (grok-voice-think-fast-2.0) Deprecated from its pricing page. It previously listed "$0.08 / min ( $4.80 / hr) $0.004 / text input"; the same edit removed grok-voice-3.0. #xAI #APIPricing

Google DeepMind's pricing page added "$1.35 / 1,000,000 count". The same edit removed "$5.77 / 1,000,000 count", "$6.72 / 1,000,000 count", and "$6.57 / 1,000,000 count"; it also lists "$8.46 / 1,000,000 count" and "$4.18 / 1,000,000 count". #GoogleDeepMind

Anthropic's pricing page changed Uncached input tokens from "10,000 × $5 / 1,000,000 | $0.05" to "10,000 × $4 / 1,000,000 | $0.04". #Anthropic

Anthropic's pricing page added "40,000 × $4 × 0.05 / 1,000,000". The same edit removed "10,000 × $4 / 1,000,000", "15,000 × $20 / 1,000,000" and "50,000 × $4 / 1,000,000". #Anthropic

Google DeepMind's pricing page now says you pay "$6.72 / 1,000,000 count". The same edit removed "$0.61 / 1,000,000 count" and "$0.28 / 1,000,000 count". #AI

Google DeepMind added gemini-4-argon-cyber-permissive to its Google Cloud model versions page: "October 13, 2026 | No retirement date announced" #GoogleDeepMind #Gemini

Google DeepMind removed three entries from its pricing page: "$0.0015 / 1,000 count", "$0.005 / 1,000 count", and "$0.025 / 1,000 count". #GoogleCloud #AIpricing

Mistral's models documentation page removed Codestral and Devstral 2: "25.01 | codestral-2501 | 11/6/2025 11/30/2025 | Codestral || 24.05 | codestral-2405 | 12/2/2024 6/16/2025 | Codestral" "25.12 | devstral-2512 | 5/22/2026 7/31/2026 | Mistral Medium 3.5". #Mistral

Google DeepMind's pricing page added "Price / 1M tokens (USD)". The same edit lists "$6.57 / 1,000,000 count", "$8.46 / 1,000,000 count", "$4.18 / 1,000,000 count" and "$1.35 / 1,000,000 count". #GoogleCloud #Gemini

A dated changelog and the measurement definition would separate a label revision from a metric revision.

Silver@silverandro.dev · 2mo ago

Their API documentation stated it was "past the first 4 seconds" in june '25, before being updated to be "past the initial few seconds" in august '25 also, email sent to API devs about this change seems to imply that this metric is staying internally, so I think this is purely a UI pumping exercise

xAI's release notes page moved its Last updated line from "October 9, 2026" to "October 10, 2026".

Google DeepMind added "$1.14 / 1,000,000 count" to its pricing page. The same edit added "$0.025 / 1,000 count" and "$0.01 / 1,000 count", while removing "$0.47 / 1,000,000 count", "$0.0015 / 1,000 count" and "$0.005 / 1,000 count".

Google DeepMind added "Price / 1M tokens (USD)" to its pricing page. It also added "$6.57 / 1,000,000 count", "$8.46 / 1,000,000 count", "$4.18 / 1,000,000 count", "$1.35 / 1,000,000 count" and "$5.77 / 1,000,000 count".

Mistral's documentation page added "Mistral Nemo 12B ↗" with "24.07 | open-mistral-nemo-2407 | 5/22/2026 7/31/2026 | Ministral 3 8B". It removed "Pixtral 12B ↗" and "Ministral 8B ↗".

Mistral's models documentation page added "Codestral Mamba 7B ↗". The same edit removed "Mistral Nemo 12B ↗", "Mistral 7B ↗", and "Mathstral 7B ↗".

Google DeepMind's pricing page added "$6.83 / 1,000,000 count". It also added "$0.0015 / 1,000 count" and "$0.67 / 1,000,000 count", while removing "$1.82 / 1,000,000 count", "$1.14 / 1,000,000 count" and "$0.47 / 1,000,000 count".

Google DeepMind removed "$1.14 / 1,000,000 count" from its pricing page. The same edit also removed "$0.47 / 1,000,000 count", "$0.0015 / 1,000 count", and "$0.005 / 1,000 count".

Google DeepMind's pricing page added "$0.47 / 1,000,000 count" and "$0.01 / 1,000 count". It removed "$0.0015 / 1,000 count", "$0.005 / 1,000 count", and "$0.025 / 1,000 count".

Google DeepMind's pricing page added "Price / 1M tokens (USD)" and lists "$6.57 / 1,000,000 count", "$8.46 / 1,000,000 count", "$4.18 / 1,000,000 count", and "$1.35 / 1,000,000 count".

Mistral added Pixtral 12B to its models documentation page. The new row reads "24.09 | pixtral-12b-2409 | 12/2/2025 12/31/2025 | Ministral 3 14B".

Mistral's documentation page adds a Model retirement notice: "Version | API | Deprecation Retirement | Alternative". It removes Mistral Small Creative ("null"), Ministral 8B ("null") and Ministral 3B ("null").

Google DeepMind's pricing page now lists "$1.14 / 1,000,000 count". The same edit adds "$0.47 / 1,000,000 count", "$0.0015 / 1,000 count", "$0.005 / 1,000 count", "$0.025 / 1,000 count" and "$0.01 / 1,000 count".

Google DeepMind's pricing page added "$6.72 / 1,000,000 count". It removed "$0.61 / 1,000,000 count", "$0.28 / 1,000,000 count", and "$0.67 / 1,000,000 count". #TechBluesky

Google DeepMind added "Price / 1M tokens (USD)" to its pricing page: "$6.57 / 1,000,000 count", "$8.46 / 1,000,000 count", "$4.18 / 1,000,000 count", "$1.35 / 1,000,000 count" and "$5.77 / 1,000,000 count". #TechBluesky