Google made Gemini 4 Carbon generally available through its API and AI Studio, while Anthropic's Fable 5.2 keeps the lead on FrontierCode and FrontierSWE. Anthropic has also launched its IPO roadshow.
Koda editorial team · 11 November 2026 · 7 min read
Listen · 16 min
Today: Test Carbon on your own agent workload this week. The steps
Google made Gemini 4 Carbon generally available this week through its API and AI Studio, at a reported $2 per million input tokens and $10 per million output tokens2; consumer access is limited to AI Ultra subscribers. Carbon reportedly leads Zapier AutomationBench and, among the models Google compared, Terminal-Bench 4.0, while Anthropic's Fable 5.2 leads FrontierCode and FrontierSWE.
Why it matters. Developers building agentic automation can now buy Carbon through the API and AI Studio at $10 per million output tokens2.
Go further: today's paper, How are viral trends different from educational and commercial uses of LLM image generation and editing? A privacy-preserving descriptive analysis (arXiv, 5 Nov).
Google researchers used Gemini 4 Carbon with other models to formally verify key security properties in code behind Signal's encryption protocol. The work reportedly took weeks.
Why it matters. Budget AI-assisted formal verification for your own cryptographic code only once Google publishes which properties were proven and how much human review the proofs needed.
More than 6,000 prebuilt AI agents and assets from IBM Consulting anchor a new industry-specific agent platform IBM and Microsoft announced on November 10.3
On BioBasicsBench, 100 routine lab-science tasks from OpenAI and Ligo, GPT-5.6 Sol earned 43.8%4 of available credit; GPT-6 Astra now earns 66.7%.
Why it matters. Read 66.7%4 as partial credit across 100 tasks, and check which protocol steps Astra still fails before letting it draft lab procedures unsupervised.
The Information reports that Nvidia CEO Jensen Huang's years-long ties to Alice Huang, described as a suspected spy, helped move more than $7.5B1 in Nvidia chips into China. The report adds to an FBI probe that President Trump has confirmed, a shareholder lawsuit and a $190B drop in Nvidia's market value.
Anthropic released Fable 5.2 and Mythos 5.2, which share one underlying model; Fable 5.2 adds extra protective guardrails. Anthropic pitches Fable 5.2 for its users' most ambitious projects, says it works up to 30%5 faster than before, and calls it a "step above" Opus 5.5.
Faire retrained engineers as "product builders" who combine coding with product and design thinking, letting them run features end-to-end. That leaves PMs and designers free for complex customer problems, and early results showed engineers contributing more to customer outcomes.
A December Fed hike went from possible to expected after October U.S. inflation beat forecasts. Dallas Fed President Lorie Logan said at least two more quarter-point hikes may be needed.
Why it matters. Each added hike raises borrowing costs for debt-funded data center and GPU buildouts, so AI capex plans priced on cheaper money need a fresh look.
The Nasdaq fell 0.74% on the 10 Nov session, with the S&P 500 and WTI crude each down 0.38%. On 11 Nov, Ether dropped 4.02% and Bitcoin 2.41%, and the crypto Fear & Greed index reads 29 (Fear).
S&P 500
7,520.93
-0.38%
Nasdaq
25,065.88
-0.74%
Bitcoin
$73,408.65
-2.41%
Ethereum
$2,045.83
-4.02%
Crude Oil (WTI)
$86.52
-0.38%
Crypto Fear & Greed
29 / 100
Fear
Levels from Yahoo Finance6, 10 Nov and 11 Nov; alternative.me7, 11 Nov.
Scheduled automation is what separates Happycapy from a chat window: an agent works on a timer inside hosted compute and mails its findings to a Capymail address, with no local environment to build. Product Hunt reviewers like that execution-first feel, though they also report thin visibility into long jobs and scheduled runs that missed while the machine slept. Non-developers who want recurring research or monitoring have a reasonable bet in a month on Pro; terminal-native developers lose little by staying on Claude Code.
The price is posted, so test before you commit.
Carbon's general availability gives agent builders a new option, and Fable 5.2 still leads the coding benchmarks, so the right model depends on your workload. Our deep dive on Anthropic's reported $3.4B9 gas generator purchase argues that the Caterpillar deal buys four years of power, but a 2.4x ramp in yearly compute obligations suggests API prices may stall through 2028. Take that as a reason to budget on today's rates and not wait for cuts. Run your own task on both models this week and pick on measured results.
Below, by source, are the 15 figures we checked: 3 verified, 8 reported and 4 estimated.
Verified confirmed by an independent searchReported as published by the named sourceEstimated approximate, in the source or in our check
Signal 229 · 11 November 2026 · 12 sources · How every number is checked · Corrections
228 editions so far, one a day. Unsubscribe anytime.
Forward this to one operator you work with. Your referral link is in every email; milestones at koda.community/refer.