Pondero AI logo

Pondero AI

Archives
Log in
Subscribe
July 16, 2026

Pondero Brief: 2026-07-16: Murati's lab ships a 975B open-weight model, free to download

Pondero Brief - JULY 16TH, 2026

Plus a $1.5B Anthropic services JV and DeepSeek's 8-day API migration deadline.
pondero. BRIEF
Thinking Machines Lab releases Inkling: 975B open-weight model
MODEL RELEASE

Mira Murati's Thinking Machines Lab ships Inkling: 975B parameters, open weights, free to download

The 41B active footprint is what makes a frontier-scale open model runnable off-cloud, and Apache/MIT weights point to Cursor, Continue, and n8n integrations within weeks.

JULY 16TH, 2026 · BY JONATHAN HILDEBRANDT

Mira Murati's Thinking Machines Lab released Inkling on July 15, its first public model since the lab launched in late 2024. Inkling is a 975B-parameter mixture-of-experts model that fires only 41B parameters per token, ships a 1M-token context window plus native audio, and comes with full weights under an Apache/MIT license, per the lab's launch post.

Why it matters. The 41B active footprint is what makes a frontier-scale open model runnable off-cloud, and Apache/MIT weights point to Cursor, Continue, and n8n integrations within weeks. Self-hosted coding agents now have an open baseline to beat.

Read our writeup →
 
Money & Moves
Ode with Anthropic enterprise AI services firm launch

Ode with Anthropic officially launches as a new enterprise AI services firm

JULY 16TH · PONDERO NEWSDESK

Anthropic, Blackstone, and Hellman & Friedman launched Ode with Anthropic on July 15, an independent enterprise AI services firm capitalized at $1.5 billion with Goldman Sachs also backing it, per TechCrunch. It runs exclusively on Claude, opened with 100 engineers, and targets healthcare, financial services, and industrial clients. Anthropic holds a minority stake.

Why it matters. Anthropic is betting the next enterprise AI money is in implementation, not model training, and it now has a named vendor squaring off against Microsoft Frontier and Accenture AI. Any company already on Claude Max that wants dedicated build engineering has somewhere to send the RFP. Read our take →

 
Tools & How-To
Beehiiv Group Subscriptions feature launch

Beehiiv shipped Group Subscriptions

JULY 16TH · PONDERO NEWSDESK

Newsletter operators can now sell team access from one checkout, with an admin dashboard for seat management, live today on the Max plan ($109/month, or $96/month billed annually), per Beehiiv. If you sell a newsletter into companies, this ends the manual multi-seat invoicing. Read our writeup → · Try Beehiiv →

 
Quick Hits
• DeepSeek's old API endpoints retire July 24. The deepseek-chat and deepseek-reasoner strings return hard 404s after 15:59 UTC on July 24, per DeepSeek's changelog - swap to deepseek-v4 or v4-pro now. V4-Pro also stacks 2x peak-hour pricing from 1 to 10 AM UTC. Migration note →
• WAIC 2026 opens July 17 with Xi's first keynote. Huawei's Atlas 950 SuperPoD (8,192 Ascend 910D chips) and 300-plus product debuts land in Shanghai; the chip-supply story is the one to watch. Roundup →
• Microsoft patched a record 570 CVEs on Patch Tuesday. Nearly triple June's count; two actively exploited zero-days hit AD Federation Services and SharePoint. Patch those two first. Details →
From the Pondero Stack
Claude Fable 5 vs GPT-5.6 Sol comparison

Fable 5 is free through July 19, then Sol undercuts it. Pick now.

JULY 14TH · PONDERO COMPARISON

Claude Fable 5 is free until July 19, then $50 per million output tokens; Sol runs $30 - per our pricing breakdown. For a heavy-output coding loop, lock your default before the free window closes. Read it → · Try in Cursor →

 
Cline ClinePass review

ClinePass earns its monthly fee once you outgrow BYOK.

JULY 15TH · PONDERO REVIEW

ClinePass adds open-weight model access for $9.99/month to a tool that was already free, per our review. For steady daily coding it beats bring-your-own-key billing; for occasional use, stay on BYOK. Read it → · Try it →

 
MCP vs A2A agent protocol guide

MCP and A2A are not rivals. Start with MCP.

JULY 15TH · PONDERO GUIDE

Despite the July 14 rival framing, MCP wires an agent to its tools and A2A wires one agent to another. Build the tool layer with MCP first, add A2A only once you have multiple agents to coordinate. Read it →

 
GitHub Agentic Workflows security guide

GitHub Agentic Workflows has an unpatched exposure. Audit now.

JULY 12TH · PONDERO GUIDE

The GitLost issue had no patch as of July 12, so the fix is your own permission hygiene. Our seven-step audit cuts what an agentic workflow can reach inside your repo. Read it →

 
Lumo vs ChatGPT private AI comparison

For data you cannot send to US clouds, Lumo 2.0 clears the bar.

JULY 13TH · PONDERO COMPARISON

Lumo 2.0 posted its first benchmark worth arguing about. For compliance teams that cannot share data with US providers, it is the pick; everyone else stays on ChatGPT. Read it →

 

How was today's brief?

★★★★★ Nailed it  |  ★★★ Solid  |  ★ Missed

Some links in this brief go to affiliate partners. Pondero may earn a commission if you click and purchase, at no extra cost to you. See our full affiliate disclosure.

Jonathan Hildebrandt Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.

X / Twitter  ·  Bluesky  ·  LinkedIn

Affiliate disclosure  ·  Unsubscribe  ·  Manage preferences

Pondero earns commissions on some links. This does not affect our editorial picks.

Don't miss what's next. Subscribe to Pondero AI:
← Newer Pondero Brief: 2026-07-17: Anthropic's $1.5B Ode bets on implementation, not models Older → Pondero Brief: 2026-07-15: Copilot's /security-review lands in-app for every plan tier
pondero.ai
Bluesky
LinkedIn
Twitter
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.