Model updates, agent tooling, and AI workflows #11

Model updates, agent tooling, and AI workflows #11

Today's Letter

  1. OpenAI, GPT‑5.5 Instant set as ChatGPT default
  2. Google, Gemma 4 MTP drafters for faster inference
  3. Anthropic releases finance agent templates
  4. GitHub Copilot commit co-author default update

OpenAI, GPT‑5.5 Instant set as ChatGPT default

OpenAI, GPT‑5.5 Instant set as ChatGPT default
  • OpenAI updated ChatGPT’s default model to GPT‑5.5 Instant on May 5, 2026, positioning it as a smarter and more accurate daily-use model.
  • The company said internal evaluations showed 52.5% fewer hallucinated claims than GPT‑5.3 Instant on high-stakes prompts in medicine, law, and finance.
  • OpenAI also reported a 37.3% reduction in inaccurate claims on difficult conversations that users had previously flagged for factual errors.
  • The release adds improvements in image and photo analysis, STEM question answering, and model decisions about when to invoke web search.
  • OpenAI said responses are now shorter and clearer, with less overformatting, fewer unnecessary follow-up questions, and less gratuitous emoji use.
  • The post also emphasizes stronger use of previously shared context for personalization, with new controls around how responses are tailored.
  • GPT‑5.5 Instant is available as the default ChatGPT model for all users, according to the release.

Source: openai.com
More: news.google.com


Google, Gemma 4 MTP drafters for faster inference

Google, Gemma 4 MTP drafters for faster inference
  • Google introduced Multi-Token Prediction drafters for Gemma 4, positioned as an inference-speed upgrade for the model family
  • The company says the approach can deliver up to a 3x speedup, targeting generation latency rather than a new model release
  • The method adds lightweight drafters that predict multiple upcoming tokens, while the main Gemma 4 path verifies and advances decoding from those candidates
  • Google highlighted integration paths across LiteRT-LM, MLX, and Hugging Face Transformers, which points to both on-device and developer-stack use cases
  • Gemma 4 31B is part of the supported lineup mentioned in the release materials, alongside the broader Gemma 4 ecosystem
  • Google says Gemma models were downloaded more than 60 million times in the first few weeks, giving the optimization immediate relevance for existing deployments
  • The update focuses on inference bottlenecks and throughput efficiency, which may reduce serving cost or improve responsiveness for apps built on Gemma 4
  • Google published the announcement on 2026-05-05 through its official blog, and the core speedup and rollout details align with supporting coverage

Source: blog.google
More: news.google.com


Anthropic releases finance agent templates

Anthropic releases finance agent templates
  • Anthropic introduced ten ready-to-run agent templates for financial services work, including pitchbooks, KYC screening, valuation review, and month-end close tasks.
  • The templates ship as plugins for Claude Cowork and Claude Code, and as cookbooks for Claude Managed Agents on the Claude Platform.
  • Each template packages task-specific skills, governed data connectors, and subagents for focused steps such as comparables selection or methodology checks.
  • Anthropic said the release is paired with Claude Opus 4.7, which it says leads Vals AI's Finance Agent benchmark at 64.37%.
  • Claude add-ins for Microsoft 365 now extend the workflow into Excel, PowerPoint, and Word, with Outlook listed as coming soon.
  • Anthropic said context can carry across those apps, so work started in a model can continue into decks or documents without re-entering the same instructions.
  • New connectors add governed real-time access to providers including Dun & Bradstreet, Financial Modeling Prep, Guidepoint, IBISWorld, SS&C IntraLinks, Third Bridge, and Verisk.
  • Moody's also launched an MCP app that brings proprietary credit ratings and company data covering more than 600 million public and private companies into Claude.

Source: anthropic.com
More: news.google.com


GitHub Copilot commit co-author default update

  • GitHub posted an update on Copilot commit attribution in VS Code issue #314311.
  • VS Code 1.110 added `git.addAICoAuthor` for `Co-authored-by: Copilot <[email protected]>` in commit messages.
  • The setting supports `off`, `chatAndAgent`, and `all`.
  • In 1.110, the default value was `off`.
  • In 1.117, the default changed to `all`.
  • Public rollout of 1.117 started on 2026-04-22.
  • The May 5 issue says the 1.117 rollout included a bug in this attribution flow.
  • Users can switch the setting to `off` or `chatAndAgent` to limit automatic attribution.

Source: github.com
More: news.google.com


Jocoletter curates AI, software, and product trends for developers and builders.

#Anthropic #GitHub #Google #OpenAI

Subscribe to Jocoletter

Read more