AI Developer Updates: New Models, APIs & Breaking Changes
Résumé
Pixel’s announcement reviews AI developer updates from Nov 9–16, 2026, including Claude Fable 5.2, Gemini 4 Carbon, decision models, and DeepSeek’s application API suite. It highlights pricing, integration changes, and potential breakages, plus six Copilot models scheduled for removal on Dec 14.
DEV Community
AI updates for developers, Nov 9–16 2026: Fable 5.2, Gemini 4 Carbon, decision models, and what breaks
By the Cuelark Team. Every release, price and date below comes from the vendor's own blog, docs or changelog, checked on 16 November 2026. The Reddit section links every thread we used. This week was busy for anyone shipping on top of model APIs. Anthropic and Google both put out new frontier models. Two vendors now sell "decision" models that answer closed questions instead of writing text. DeepSeek turned its API into a full application stack. And GitHub put a date on retiring six Copilot models. Here is what changed, what it costs, and which parts can break existing code. We covered Gemini 4 Hydrogen separately in our React chat migration post, so it only comes up here where it touches other changes. TL;DR Date (2026) What Why a developer should care Nov 9 Claude Fable 5.2 and Mythos 5.2 $8/$40 per 1M tokens. Preserved thinking now applies to all accounts, which can 400 your requests if you edit history Nov 9 Meta Muse Ember on the Meta Model API $0.10 / $0.50 per 1M tokens, also on Google Cloud, Oracle Cloud and Vercel AI Gateway Nov 10 Gemini 4 Carbon generally available to developers Intro price $2/$10 until Nov 30, then $4/$20. Sampling params ignored, explicit caching unsupported Nov 11 Gemini Decisions API (gemini-decision-1) Up to 1,000 yes/no, choice or rubric questions per call. $0.10 per 1M input, output free Nov 11 Qwen4-Max-Preview Two-week preview on Token Plan only. max_tokens now covers thinking plus answer Nov 12 Inkling 2 Small weights 594B total / 17B active MoE, Apache 2.0, runs in vLLM and SGLang Nov 13 DeepSeek application API suite Built-in web search, URL reading, file input, OCR, ASR, TTS. A safety parameter required for public apps Nov 13 GitHub Copilot deprecations Six models removed from Copilot on Dec 14 Claude Fable 5.2 and Mythos 5.2: cheaper, faster, and one API change to check Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9. Price: $8 per million input tokens and $40 per million output, lower than 5.1. An optional fast mode generates up to 2.5× faster for $16/$80. Speed: output is up to 30% faster than 5.1. Availability: Fable 5.2 is on every platform, including AWS, Google Cloud and Azure, as claude-fable-5-2. Mythos 5.2's cyber capabilities go through Anthropic's Cyber Verification Program. Which model to start with: Anthropic still tells most teams to start on Opus 5.5 and move up to Fable only for the hardest work. Its own numbers are the main reason to look anyway: on FrontierCode 2.0, Fable 5.2 at High effort scored about level with Opus 5.5 at Max effort (45.3% vs 45.5%) for roughly half the cost per task. That's Anthropic's benchmark, not ours. Advisor and orchestrator roles: the post describes using Fable 5.2 as an advisor that cheaper Haiku, Sonnet or Opus models consult, or as an orchestrator that hands simpler work down to them. Deliberate limits: during training, Anthropic took steps to hold two research skills, accelerator kernel optimisation and improving LLM architectures, at about Opus 5.5's level. It says agentic coding and general ML work weren't affected. The breaking change: preserved thinking for everyone One line in the launch post matters more for integrations than any benchmark: "preserved thinking now applies to all accounts". Anthropic frames it as an anti-distillation measure. The preserved thinking docs explain the practical side: A thinking block is bound to everything before it: system, tools and every earlier message. If any of those change between requests, the block fails a prefix check and the request can get a 400 error. prefix_mismatch_behavior lets you choose between that 400 and having the API drop the failing blocks. The response's input_transformations field tells you which blocks were dropped. thinking: {"type": "enabled"} is now a 400 on these models. Send "adaptive" or leave thinking out. Thinking blocks from Fable 5.2, Mythos 5.2, Sonnet 5.5 and Haiku 5.5 only work in the account (or linked account) that produced them. Anywhere...

Powered by Readybot.io