Skip to content
Agentdex

Model Pulse #17 · Tuesday, Nov 17, 2026

Isomorphic Labs raises $5B, Model ML hits a $1.5B valuation, Alibaba ships Qwen4-Max, and Unity's CLI hits 1.0

The six AI stories since Saturday's issue that matter most for AI coders and indie game devs. Each one links to its primary source.

Sponsored Sponsor this issue: reach AI coders and solo founders in Model Pulse.

Nov 17, 2026

DEALS: Isomorphic Labs raises a $5B Series C co-led by Greenoaks and Sequoia, and Model ML raises $250M at a $1.5B valuation

Isomorphic Labs, the Alphabet-backed drug-design company run by Demis Hassabis, announced a $5B Series C on Nov 17. Greenoaks and Sequoia Capital co-led the oversubscribed round, with Andreessen Horowitz and other new investors joining, and what the company called continued significant investment from Alphabet. All existing investors took part: Thrive Capital, MGX, GV, CapitalG, Temasek and the UK Sovereign AI Fund. The company did not disclose a valuation. The money is meant to scale IsoDDE, its drug-design engine of simulation models and agents, and to build translational and clinical infrastructure as programmes move toward trials. Isomorphic says it is seeing positive preclinical results in its own oncology and immunology programmes and with partners (company claims). Hassabis told Bloomberg, as reported by The Next Web, that more than 20 programmes are underway, that partners include Eli Lilly, Novartis and Johnson & Johnson, and that the new capital will speed the move toward clinical trials. TNW notes the round comes six months after a $2.1B Series B in May. Also today, Model ML announced a $250M Series B at a $1.5B valuation, 18 months after coming out of stealth. General Atlantic and QED Investors co-led, taking total funding to $350M. The London-headquartered company, founded by brothers Chaz and Arnie Englander, builds a personal AI agent for bankers, consultants and investors that runs on a firm's own systems and data. It says customer data is never used to train models, and that it often deploys inside the customer's own infrastructure. It is ISO/IEC 27001:2022 certified and SOC 2 Type II audited, and is not tied to one model provider. In the third quarter of 2026, it says annual recurring revenue grew 12x year on year and daily active users 26x (company figures).

Source: https://www.isomorphiclabs.com/articles/isomorphic-labs-announces-series-c-investment-round

Nov 16, 2026

MODELS: Alibaba releases Qwen4-Max, a 2.4-trillion-parameter MoE with a 1M-token context, at $1.50/$4.50 per million tokens, and keeps the weights closed

Alibaba released Qwen4-Max on Nov 16, calling it the most capable model in the Qwen family so far. It is a sparse mixture-of-experts model with 2.4 trillion total parameters that activates 71 billion per token, 25% fewer than Qwen3.8-Max at the same 2.4T scale, according to Alibaba. It reads text, images and video, has a 1-million-token context, and is available as qwen4-max on QwenCloud, with OpenAI- and Anthropic-compatible APIs, plus QwenWork and QwenStudio. QwenCloud's model page lists Singapore/international prices of $1.50 per million input tokens and $4.50 per million output tokens (thinking tokens count as output), with implicit or explicit cache reads at $0.075. A limited preview opened on Nov 11. Alibaba says the model ranked 6th on Code Arena, and that on its own Work360 test it beat Qwen3.8-Max overall while using 35% fewer tokens at High effort and cutting average cost per task by around 60% (company figures). The Next Web, reading charts Alibaba published, notes a mixed picture: 75.7 on DeepSWE v1.1, just ahead of Claude Opus 5.5 at 74.2, but 48.5 on Terminal-Bench 4.0, behind Opus 5.5 at 59.6 and Kimi K3.1 at 67.2. The weights are closed. Alibaba says smaller Qwen4 models will get open weights, with no date given.

Source: https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max

Nov 17, 2026

GAME DEV: Unity's CLI leaves beta at Unite Tokyo as version 1.0, with official plugins for coding agents and a promise of 1.x compatibility

Unity released the Unity CLI and the Pipeline package (com.unity.pipeline) as version 1.0 on Nov 17 at Unite Tokyo, ending a beta that started on July 15. The CLI installs and manages Editors, modules and projects from the terminal, and the Pipeline package lets a script or coding agent drive a running Editor: open a project, run a command, enter Play mode and read the result. Unity says that since July 20 more than 350,000 logged-in users have run more than 600 million commands, and that about 98% of those runs come from coding agents and other automation (company figures). unity setup installs the official plugin for Claude Code, Codex, Grok Build or Google Antigravity, and unity skill install --all adds Unity's own skills to every agent on the machine, including VS Code, GitHub Copilot, Qwen Code and OpenCode. The skills are also public on GitHub. Version 1.0 can see and answer a hidden Editor dialog so an agent doesn't hang, builds through the Editor you already have open, writes stable JSON output, never prompts in CI, and can pin a CLI version range per project. Unity also says commands, flags and output will stay compatible across 1.x releases, with a deprecation warning before anything is removed. The Build Server license now covers agents and automation, one license per instance, including a headless Editor or a container. A supported Linux containerized Editor is listed as coming soon.

Source: https://unity.com/the-unity-cli-1.0-is-here

Nov 16, 2026

OPEN SOURCE: TokenEconomizer, a plugin from UC Berkeley and Microsoft Research, cuts coding-agent costs by learning from past sessions in the same repo

Microsoft Research, Microsoft Security and UC Berkeley published TokenEconomizer on Nov 16. It is an open-source plugin that sits beside a coding agent you already use and, after each session, updates what it knows about that repository. It does not change the agent or the model. Three tricks do the work: micro-skills, one-line notes that are added to the prompt only after several sessions confirm them; tool pruning, which hides tools the repo rarely uses while always keeping shell, read and edit; and a failure-loop check that suggests a different approach when the agent is stuck, then alerts you rather than killing the job. On 277 SWE-Bench bug-fixing tasks across six repositories and five agent-and-model combinations, the authors report 18.4–30.9% lower mean cost per task, cheaper results in all 30 repository-combination pairs, and a higher Pass@1 score in four of the five combinations. Savings include the plugin's own model calls, and every repository was cheaper than the default by the sixth task. They also report 16.0–21.2% lower costs on Web-Bench. The plugin is on GitHub, works with Claude Code, Codex, the GitHub Copilot CLI and Pi, needs no API key of its own, and keeps what it learns on your machine. The numbers are from the authors' own paper, arXiv 2610.22732.

Source: https://www.microsoft.com/en-us/research/blog/tokeneconomizer-learning-from-experience-to-make-ai-coding-agents-more-efficient/

Nov 16, 2026

CODING TOOLS: JetBrains' experimental TeamCity plugin runs Codex and Claude Code inside CI builds, and wipes provider keys when the build ends

JetBrains released an experimental AI Agents plugin for TeamCity on Nov 16. When a build starts, the plugin checks whether Codex or Claude Code is on the build agent and installs it if needed, so you call codex or claude from a normal Command Line build step. Provider credentials live in a TeamCity connection, not on the machine. The plugin supports OpenAI, Anthropic, OpenRouter, LiteLLM and JetBrains Air, hands the keys to the agent only for the length of the build, and removes them afterward. Each run publishes prompt, completion, cached and total token counts as TeamCity statistics, and saves the agent's trajectory as HTML for reading and JSON for analysis, in hidden artifacts under .teamcity/ai-agents/. JetBrains warns that those files may contain prompts, source code and model output, so access should be restricted like any other sensitive build artifact. The plugin is experimental and may change. You install it from JetBrains Marketplace.

Source: https://blog.jetbrains.com/teamcity/2026/11/run-ai-agents-in-teamcity/

Nov 16-17, 2026

AGENTS: OpenCode Browse lets coding agents use your real signed-in browser, and hands the tab back for logins, 2FA and payments

OpenCode published Browse on Nov 16–17, an MCP server and Chromium extension that lets a coding agent work in the browser you already use, with your tabs, cookies, logins and extensions. Agents open tabs in collapsed background groups, then read pages, click, type, take screenshots, record tabs and inspect network traffic. When a login, 2FA prompt or payment appears, the agent hands the tab back to you and waits until you click Continue. You can also lend an agent a tab you already have open, and you can limit an agent's tabs to a list of allowed sites. It works with Chrome, Edge, Brave, Arc, Helium, Vivaldi, Opera and Chromium, and with MCP clients including OpenCode, Claude Code, Cursor, Codex and Zed. No API keys and no cloud service sit between the agent and the browser; the stack runs locally. The code is on GitHub under MIT. The useful part for AI coders is testing a deployed app or filing a ticket while already signed in. The risk is the same as lending someone your logged-in browser, so start with the allowed-sites list and keep payments and 2FA as handoffs.

Source: https://opencode.ai/product/browse

Ship faster this week

Grab the free Agentdex starter kit with prompts and agent configs for Cursor, Claude Code and Codex.

Get the free kit

Model Pulse

Daily AI model drops, tools, and what they mean for AI coders.

We store your email with Resend and use it only for Model Pulse. See privacy. Missed the last issue? Read Model Pulse #16.