AI News — October 17, 2026

Today's AI news in short: 6 items, each with a link to its source.

Anthropic IPO marketing timeline reportedly shifts to mid-October

According to a secondary report carried by Reuters, the Anthropic IPO marketing phase has moved to mid-October. Valuation and S-1 figures conflict across sources, so they should not be relied on.

Source: Let's Data Science

Claude Code 2.1.302 release unconfirmed; official changelog conflicts with third-party listing

No source was found for a Claude Code 2.1.302 release. A third-party page lists v2.1.296 (9 October) as the latest version, while the official changelog shows 2.1.301, so the sources conflict. The official changelog should be checked directly.

Source: Claude Code Docs

MCP specification update reportedly removes protocol-level sessions and initialize handshake

A third-party commentary says the 2026-07-28 MCP specification removes protocol-level sessions and the initialize handshake, and makes server/discover mandatory. The official specification still needs to be verified.

Source: DEV Community

Google's Gemini 4 reportedly nearing release, possibly earlier than year-end

According to The Information, Google is nearing the release of its flagship Gemini 4 model. DeepMind head Koray Kavukcuoglu reportedly hoped it would launch before the end of the year. Earlier digests named it 'Gemini 4 Argon', which does not match, so the model name and date should be checked on the official blog.

Source: Sequenzy

Hugging Face trending list features embeddinggemma-2 and Unlimited-OCR models

On the live trending list as of 17 October, google/embeddinggemma-2 (feature-extraction, dated 6 October, about 151 thousand downloads) and baidu/Unlimited-OCR (image-text-to-text, dated 14 October, about 1.26 million downloads) appear. Whether either fits on a laptop has not yet been checked.

Source: Hugging Face

Hugging Face trending list includes LiquidAI LFM2.5-24B-A2B and OpenBMB MiniCPM-V-4.7

The trending list includes LiquidAI/LFM2.5-24B-A2B (text-generation, dated 13 October) and openbmb/MiniCPM-V-4.7-35B-A3B (image-text-to-text, dated 15 October). The digest says the GGUF Q4 file size of the 35B MoE model is at the limit for 32 GB RAM, which has not yet been verified.

Source: Hugging Face

Summaries are condensed from the linked sources; follow each link for the full story.