● Biggest: Gemini 4 Hydrogen launched at $1.50/M and outscores Google’s own flagship on the AA Intelligence Index.
● Best science: AlphaGenome 2 roughly doubles causal variant accuracy by conditioning on one cell type at a time.
● Most useful if you build agents: Anthropic found per-user agent teams lose to one shared coordinator by up to 45 points.
● Quietest: OpenAI open-sourced the Codex Security CLI — the client only.
Four stories worth your time from 11 to 13 November 2026. Every figure below was read off the primary source, not a roundup.
Gemini 4 Hydrogen beats Argon at a fifth of the attention
Google shipped Hydrogen on 12 November at $1.50 input and $7.50 output per million tokens, flat across a full 1M context. It takes 51 on the Artificial Analysis Intelligence Index against Argon’s 50, triples 3.8 Flash on Terminal-Bench 4.1 at 57% versus 18.2%, and runs 2.5x faster at default. Free users get it in Google Search globally. Argon still leads on Chartography, GraphWalks and SEC-bench Pro. Full breakdown.
AlphaGenome 2 reads DNA one cell type at a time
DeepMind’s new DNA sequence model, out 11 November, trained on over 107 million individual cell profiles across 68 tissues. On the TenK10K immune-cell set it identifies causal variants 30% of the time against 16%, and gets gene-effect direction right 43% against 23%. Free for non-commercial research; explicitly not validated for clinical use. What changed.
Anthropic: one agent each is worse than one agent between you
Anthropic Fellows tested Claude Opus 5 and Sonnet 5, GPT-5.6-sol and terra, and Qwen3.8-max across four shared-resource environments. A single coordinator agent beat per-user teams by 10 to 45 points. As groups grew from 4 to 16 agents, the share that ever acted fell from 66% to 10% for Sonnet 5. Three environments are released as MAMUBench. The numbers.
OpenAI open-sources the Codex Security client
The Codex Security CLI and SDK are now Apache 2.0. They scan repositories, review staged and unstaged changes before a commit, and can fail a CI check at a chosen severity. The analysis service stays OpenAI’s, and access is limited beta. What is actually open.
Common questions
Is Gemini 4 Hydrogen better than Gemini 4 Argon?
On agentic work, yes. Hydrogen takes 51 on the Artificial Analysis Intelligence Index against Argon at 50, and leads on SWE-sweep 13.5% to 4.3%, Agents’ Last Exam 71.6% to 69.0% and OSWorld-2.1 77.6% to 74.7%. Argon leads on Chartography at 71.6% to 66.6% and holds published scores on GraphWalks and SEC-bench Pro that Hydrogen has no figure for. Pick Hydrogen for tool use and speed, Argon for heavier reasoning.
How much does Gemini 4 Hydrogen cost?
$1.50 per million input tokens and $7.50 per million output, with cache hits at $0.075. Pricing is flat across the whole 1M-token context rather than stepping up at a threshold. Artificial Analysis puts cost per task between $0.29 at low effort and $3.16 at xhigh.
Can I use AlphaGenome 2 commercially?
Not yet in general. Non-commercial research access is open through the AlphaGenome API. Commercial research access is in private preview. DeepMind states the model is for research only and has not been validated for clinical or diagnostic use.
Should I give each person on my team their own AI agent?
Anthropic’s evidence says start with one shared coordinator instead. Per-user agent teams lost by 10 to 45 percentage points across four environments, agents stopped acting as the group grew, and team episodes fabricated claims about other users in over half of runs. If you do run per-user agents, appoint a team lead agent and force unread messages to be read before an agent finishes.
Sources