K Koda Intelligence
Subscribe

The Signal · Saturday 7 November 2026

Thinking Machines puts Inkling 2 Small on Tinker at 1,527 Elo

Thinking Machines Lab launched a 594-billion-parameter multimodal MoE on Tinker that scored 1,5271 Elo on AA-Briefcase. Cloudflare also moved its Workers AI cutoff to November 20.

Koda editorial team · 7 November 2026 · 6 min read

Listen · 16 min
Cover image for the 7 November 2026 Signal: Thinking Machines puts Inkling 2 Small on Tinker at 1,527 Elo

In short

Today: Make your model dependencies survive change. The steps

The lead

Model Release · VentureBeat · 06 NOV 2026

Thinking Machines Lab launched Inkling 2 Small, a 594-billion-parameter multimodal MoE on its Tinker platform. It scored 1,5271 Elo on AA-Briefcase, the same day Ant Group shipped Ling-3.2-Flash.

Why it matters. Fine-tune on Tinker only if your workload is tables, decks and memos, since Ling-3.2-Flash adds video input and Inkling's weights cannot yet leave the platform.

Go further: today's paper, CISA: Retrofitting Dense Attention Models with Chunk-Indexed Sparse Attention (arXiv, 2 Nov).

Also today

Model Release · Google · 05 NOV 2026

Gemini Omni 1.2 Flash Reaches General Availability

Google DeepMind's video model is now generally available: clips were capped at 10 seconds and now reach 15, with multi-turn editing and a reported 30%2 price cut.

Why it matters. Model the 30%2 cut against total spend per finished clip, because longer outputs can lift the bill even at a lower rate, and no independent benchmark has verified quality.

Agents · Chosun Biz · 06 NOV 2026

Kakao's Kanana-2-v-decide Routes Korean Agent Requests

On 15,9093 internal Korean questions, Kakao says Kanana-2-v-decide ranked first; it also posted 61.34 on the public Decision Index, where Nace.AI's Drex Max v1 posted 67.52.

Why it matters. Korean-market agent builders should weigh Kakao's 61.34 Decision Index score against Drex Max v1's 67.52 before routing tools through it.

Enterprise · Cloudflare · 05 NOV 2026

Cloudflare Delays Workers AI Model Deprecations by Five Days

Cloudflare's Workers AI deprecation date was November 15 and is now November 20; after that, requests to the listed models fail with HTTP 410 Gone and error code 5028.

Why it matters. Teams still calling the listed Workers AI models get a short reprieve to migrate, but any call left in place after November 20 will fail outright.

Anthropic, via Superhuman

Anthropic Adds Claude Finance Hub

Anthropic's new finance hub, Personal Finance in Claude, lets users link bank and card accounts so Claude can surface spending trends, model future scenarios and answer money questions. It cannot move, manage or invest funds, and Anthropic stresses it carries no ads or sponsored recommendations. The feature is rolling out in beta to US users on plans including Pro and Max.

anitakirkovska.com, via TLDR

Growth Teams Rebuild Around Agents

The most AI-pilled growth teams are rebuilding their go-to-market stacks around agents, turning growth into an engineering problem. Small and medium businesses use agents to recreate software their old tools no longer cover, while larger companies move slower because of contracts, integrations and operational habits.

Diplomacy · Bloomberg · 06 NOV 2026

Iran Weighs Qatari Nuclear Proposals Once Off-Limits

Iran had refused to discuss nuclear concessions. On Friday, Bloomberg reported that it will now examine Qatari proposals on them. The deal trades reopening the Strait of Hormuz for US sanctions relief.

Why it matters. Energy and shipping buyers should watch these talks closely, since a deal reopening the Strait of Hormuz could ease the supply disruption driving their costs.

In brief

Markets

The Nasdaq rose 1.09% while Bitcoin slipped 0.13% and the crypto Fear & Greed index sat at 25, Extreme Fear, so equity buying did not carry over into crypto.

S&P 500

7,376.86

+0.31%

Nasdaq

24,579.42

+1.09%

Bitcoin

$75,840.94

-0.13%

Ethereum

$2,239.59

+0.23%

Crude Oil (WTI)

$90.34

-0.20%

Crypto Fear & Greed

25 / 100

Extreme Fear

Levels from Yahoo Finance4, 6 Nov and 7 Nov; alternative.me5, 7 Nov.

Go deeper

The Koda Signal, 7 November 2026

Thinking Machines puts Inkling 2 Small on Tinker at 1,527 Elo

0:0015:36

Every episode

Tool of the day

Kiro Bots

Coding · Koda score 6.6 / 106

Kiro Bots turn a one-off coding agent into a named worker inside Kiro Crew that keeps project memory between sessions and can direct other agents. The design is sound and the entry cost is low, yet the evidence so far is launch coverage, so adoption should start as a scoped trial. Teams sitting on a dull, repeating chore like dependency bumps are the right first customers.

Capability 7.0
Ease 6.5
Value 7.0
Momentum 5.5

Also worth a look

The bottom line

Treat every model as a rented dependency.

Cloudflare pushing its cutoff to November 20 and OpenRouter routes changing price without errors show that the ground under your stack moves on someone else's schedule, and a new launch like Inkling 2 Small adds one more platform to track. Put model choice behind configuration, monitor cost and availability, and write a tested fallback for each production route. Run new agent tooling such as Kiro Bots on small, reversible work until it earns more.

  1. Grep for literal model IDs. Move every hardcoded model string into one config file and add a daily check that logs price and availability per route, so an 80% hike shows up in a dashboard instead of an invoice.
  2. Handle Cloudflare's new cutoff. After November 20, listed Workers AI models return HTTP 410 Gone with error code 5028. Catch that response in your client and fall back to a replacement model before the date arrives.
  3. Trial one Kiro Bot on a bounded task. Give it a single small project, then check after a few sessions whether it remembers decisions and delegates sensibly. Keep it only if the memory holds up.

Sources and checks

Every story links to its outlet. Below, by source, are the 11 figures we checked: 2 verified, 8 reported and 1 estimated.

  1. 1VentureBeat. It scored 1,527 Elo on AA-Briefcase, the same day Ant Group shipped Ling-3.2-Flash, 6 Nov 2026.Verified
  2. 2Google. ...and now reach 15, with multi-turn editing and a reported 30% price cut, 5 Nov 2026.Verified
  3. 3Chosun Biz. 2 figures on this page, 6 Nov 2026.Reported
  4. 4Yahoo Finance. 5 figures on this page, 6 Nov 2026 and 7 Nov 2026.Reported
  5. 5alternative.me. Crypto Fear & Greed close, Extreme Fear on the session, 7 Nov 2026.Reported
  6. 6Koda Lab report (our score). Kiro Bots, Koda score across four dimensions, 7 Nov 2026.Estimated

Verified confirmed by an independent searchReported as published by the named sourceEstimated approximate, in the source or in our check

Every figure, one by one (11)
  1. 1,527It scored 1,527 Elo on AA-Briefcase, the same day Ant Group shipped Ling-3.2-Flash.VentureBeatVerified
  2. 30%...and now reach 15, with multi-turn editing and a reported 30% price cut.GoogleVerified
  3. 15,909On 15,909 internal Korean questions, Kakao says Kanana-2-v-decide ranked first; it also posted 61.34 on the public...Chosun BizReported
  4. 67.52...Decision Index spot, where Nace.AI's Drex Max v1 posted 67.52.Chosun BizReported
  5. 7,376.86S&P 500 close, +0.31% on the sessionYahoo FinanceReported
  6. 24,579.42Nasdaq close, +1.09% on the sessionYahoo FinanceReported
  7. $75,840.94Bitcoin close, -0.13% on the sessionYahoo FinanceReported
  8. $2,239.59Ethereum close, +0.23% on the sessionYahoo FinanceReported
  9. $90.34Crude Oil (WTI) close, -0.20% on the sessionYahoo FinanceReported
  10. 25 / 100Crypto Fear & Greed close, Extreme Fear on the sessionalternative.meReported
  11. 6.6 / 10Kiro Bots, Koda score across four dimensionsKoda Lab report (our score)Estimated

Signal 225 · 7 November 2026 · 14 sources · How every number is checked · Corrections

Get the morning Signal

224 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.