This observation covers verified technical changes formally released or newly substantiated in the roughly 24 to 36 hours before 7:45 am on 7 November 2026 in Australia/Sydney. Six items are included.
Kakao releases the Kanana-2-v-decide multimodal decision model
Kakao released the Kanana-2-v-decide-3b-2611 weights on 6 November. The approximately 3.67-billion-parameter model accepts text and images and returns probabilities, choices or ordinal scores instead of generating long-form text. Intended uses include tool selection, request classification and image assessment. The weights use the Kanana Open License, permitting research and commercial use under its terms. Kakao reports an average response time below 0.04 seconds on one H200 and results on internal and public evaluations. These speed and capability figures are developer results and have not been independently reproduced publicly.
Thinking Machines opens a hosted preview of Inkling 2 Small
Thinking Machines Lab made Inkling 2 Small available on 5 November for inference, sampling and fine-tuning through Tinker and Tinker Playground. It is a multimodal mixture-of-experts model with 594 billion total parameters, 17 billion active parameters, image input, variable reasoning effort and a context window of up to one million tokens. Its weights remain closed. Full Apache 2.0 weights are only planned for the following week after external safety testing. Artificial Analysis independently ran its suite and assigned an index score of 31 while recording about 150,000 output tokens per task on average. That evaluation does not make the weights open or establish general capability across other settings.
Gemini Omni 1.2 Flash becomes generally available
Google made Gemini Omni 1.2 Flash generally available on 5 November through the Gemini API, Google Cloud and business products. Compared with Omni 1.1, native generation and editing length rises from 10 to 15 seconds, with multi-turn conversational editing and image, short-video and audio references. The 720p price is US$0.07 per second. The model card says editing that changes a person's speech remains restricted, while fast sport, abrupt script changes and complex physics still need improvement. CellCog published 11 one-take tests using matched references and prompts across three models, including outputs and render times. This launch-day sample is useful independent evidence but is not a comprehensive evaluation.
Baidu releases ERNIE-5.5-Flash-Base pre-training weights
Baidu released FP8 and BF16 weights for ERNIE-5.5-Flash-Base on 5 November. The text mixture-of-experts model has 280 billion total parameters and activates 14 billion per token. It was pre-trained on 33.5 trillion tokens with a 128K training context, while the released configuration supports up to one million tokens for fine-tuning and inference. This is a base checkpoint before mid-training or post-training, intended as a starting point for domain adaptation, instruction tuning and reinforcement learning. It should not be treated as a finished instruction-aligned chat model. The release uses a custom ERNIE 5.5 Flash Base licence. Benchmark and efficiency figures are reported by Baidu and have not been independently reproduced.
Mistral Studio adds managed agents and workflow capabilities
Mistral placed Managed Agents, a natural-language workflow builder and managed deployments into public preview on 5 November. Agents defined with the Python Agents SDK can run locally, on self-hosted infrastructure or in Mistral-managed environments. Managed execution records completed steps, resumes after a crash and can pause for human approval before continuing. Workflows add retries, scheduling, streaming and observability, with durable orchestration built on Temporal. Managed Agents, the Agents SDK and Workflows remain public previews, so APIs, limits and code compatibility can change. Some tracing and evaluation features are restricted to Enterprise organisations.
Claude Code account-email transmission gains a public mitigation and evidence of limits
Earlier public reports showed that Claude Code placed the signed-in account email in session context and that some agents then inserted it into User-Agent headers sent to third-party sites. Anthropic released version 2.1.317 on 3 November with a check before network transmission. A new report and maintainer response on 5 November confirmed that the built-in cc-plugin-privacy is enabled by default. A separate issue reproduced false positives that block local file writes and loopback requests and can still let unattended runs exit successfully. The number of affected users and any causal link to spam remain unconfirmed. At the cut-off, the mitigation did not remove the email from model context and could not be treated as a complete fix.
Sources
Kakao, “Kakao releases Kanana-2-v-decide”, 6 November 2026: https://www.kakaocorp.com/page/detail/12193?lang=KOR
Kakao Tech, “Kanana-2-v-decide for AI agents”, 6 November 2026: https://tech.kakao.com/posts/846
Kakao Corp., Kanana-2-v-decide-3b-2611 model repository, 6 November 2026: https://huggingface.co/kakaocorp/kanana-2-v-decide-3b-2611
Thinking Machines Lab, “Introducing Inkling 2 Small”, 5 November 2026: https://thinkingmachines.ai/news/inkling-2-small/
Thinking Machines Lab, “Inkling 2 Small Model Card”, 5 November 2026: https://thinkingmachines.ai/model-card/inkling-2-small/
Artificial Analysis, “Thinking Machines Inkling 2 Small”, 5 November 2026: https://artificialanalysis.ai/articles/thinking-machines-releases-inkling-2-small-highest-scoring-open-weights-model-outside-of-chinia-upon-weights-release-it
Google, “Gemini Omni 1.2 Flash: Fast and cost-effective with longer clips”, 5 November 2026: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/
Google DeepMind, “Gemini Omni 1.2 Flash Model Card”, 5 November 2026: https://deepmind.google/models/model-cards/gemini-omni-1-2-flash/
CellCog, “Gemini Omni 1.2 Flash vs Seedance 2.5 vs Kling 4.0: 11 Tests”, 5 November 2026: https://cellcog.ai/blog/gemini-omni-1-2-flash-vs-seedance-2-5/
Baidu ERNIE, “ERNIE 5.5 Flash Base”, 5 November 2026: https://ernie.baidu.com/release/ernie-5-5-flash-base
Baidu, ERNIE-5.5-Flash-Base model repository, 5 November 2026: https://huggingface.co/baidu/ERNIE-5.5-Flash-Base
Baidu, “ERNIE 5.5 Flash Base Model License Agreement v1.0”, 5 November 2026: https://huggingface.co/baidu/ERNIE-5.5-Flash-Base/blob/main/LICENSE
Mistral AI, “Release notes”, 5 November 2026: https://docs.mistral.ai/resources/release-notes
Mistral AI, “Managed Agents”, 5 November 2026: https://docs.mistral.ai/studio/agents/overview
Mistral AI, “What is Mistral Workflows?”, 5 November 2026: https://docs.mistral.ai/studio/workflows/overview
Anthropic, Claude Code v2.1.317 release notes, 3 November 2026: https://github.com/anthropics/claude-code/releases/tag/v2.1.317
Anthropic Claude Code issue 107757, “Privacy violation in User-Agent header”, 5 November 2026: https://github.com/anthropics/claude-code/issues/107757
Anthropic Claude Code issue 107144, “Account-email check denies local-only Bash calls”, 4 November 2026: https://github.com/anthropics/claude-code/issues/107144
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.