Models & Labs
Thinking Machines releases open-weight AI model Inkling
TechCrunch AIhigh confidence
Why it matters
- →Inkling offers a customizable alternative to proprietary AI models, allowing organizations to fine-tune it for specific needs.
- →The open-weight design challenges the dominance of closed models by giants like OpenAI and Google.
- →It emphasizes cost efficiency and adaptability, potentially outperforming proprietary models in specialized tasks.
Thinking Machines Lab has launched Inkling, its first open-weight AI model, which allows developers to download and modify it directly. Unlike the proprietary models from OpenAI or Google, Inkling is designed for adaptability, using a mixture-of-experts system with 975 billion parameters. The model is aimed at enterprises looking to fine-tune AI for specific tasks, offering a customizable alternative to the one-size-fits-all approach. While not claiming to be the strongest, Inkling focuses on well-rounded performance and cost efficiency, challenging the dominance of closed AI models.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1Thinking Machines Develops Real-Time AI Interaction Models
The Verge AI · May 11, 2026 · Related
- 2Parameter Golf Explores AI-Assisted Research
OpenAI · May 12, 2026 · Background
- 3Thinking Machines unveils interactive AI model
TechCrunch AI · May 12, 2026 · Related
- 4TML Unveils Real-Time AI Interaction Models
The Rundown AI · May 12, 2026 · Related
- 5Thinking Machines Labs Unveils New Interaction Models
Matt Wolfe · May 16, 2026 · Related
- 6Specialized AI Models Outperform Larger Counterparts
Hugging Face Blog · May 22, 2026 · Background
- 7Open Models Propel AI Research at ICML 2026
NVIDIA Blog · July 6, 2026 · Background
- 8Together AI Launches Provisioned Throughput for Open Models
Together AI Blog · July 8, 2026 · Background
What happened next
- 9Debate Over Open-Weight AI Models Intensifies
TechCrunch AI · July 20, 2026 · Related
- 10Together AI Enhances Open-Weight Inference Platform
Together AI Blog · July 23, 2026 · Related
- 11China's Open AI Models Challenge US Dominance
The Verge AI · July 27, 2026 · Related
- 12Chinese Labs Lead in Open Model Scale
Hugging Face Blog · August 14, 2026 · Related
- 13MIT shows AI rewriting model weights without training
The Rundown AI · November 3, 2026 · Related
Follow this story
Open the full story →Together AI launches Inkling model on day 0
5 developments
- Jul 15 · Together AI Blog
Together AI launches Inkling model on day 0 - Jul 15 · Hugging Face Blog
Thinking Machines Launches Inkling Multimodal LLM - Jul 15 · TechCrunch AI
Thinking Machines releases open-weight AI model Inkling (This article) - Jul 15 · WIRED AI
Thinking Machines Lab Releases Inkling Model - Jul 20 · Fireship
Thinking Machines releases Inkling model
More from TechCrunch AI
© TechCrunch AIAgentsagents
Meta Muse AI Agent Now Places US Business Calls
Meta’s consumer agent Muse is rolling out voice calling to US businesses for paid subscribers, targeting the tedious task of customer service navigation. With over 10 million users, Meta is betting that AI’s infinite patience will replace human frustration with phone trees and hold times. The feature leverages improved voice mimicry to interact with receptionists, though many still hang up upon detecting non-human voices. This moves Muse from text-based assistance into active, autonomous execution of real-world tasks, marking a significant step toward mainstream consumer AI utility.
TechCrunch AI
© TechCrunch AIMarket & Regulationbusiness
Nebius delays data centers, cuts revenue forecast
Nebius is facing the harsh reality of scaling AI infrastructure: even with $40 billion in planned spending and massive prepayments from Microsoft and Meta, physical construction delays are derailing timelines. A key New Jersey site is months behind schedule due to regulatory fines and generator issues, forcing a shift in delivery dates to 2027. This isn't just a logistical hiccup; it signals that the 'neocloud' model of financing builds with customer cash is fragile when third-party operators fail. The market reacted swiftly, dropping shares by 17% after Nebius cut its 2026 revenue forecast, proving that hype cannot override concrete capacity.
TechCrunch AI
© TechCrunch AIModels & Labsmodels
Google deploys Gemini 4 Hydrogen for faster Search
Google is quietly shifting its search infrastructure by replacing older fast models with Gemini 4 Hydrogen, a tier designed specifically for speed and cost-efficiency rather than raw reasoning power. This model completes tasks 2.5 times faster than its predecessor while beating the more powerful Argon on specific intelligence benchmarks, proving that specialized efficiency can outperform brute force in high-volume workflows. By integrating this into Search's fan-out technique, Google now breaks down complex queries into subqueries with greater precision and speed, delivering richer visual responses without the latency of heavier models. This marks a strategic pivot toward agentic execution layers within consumer search, making AI Mode feel snappier for everyday multi-part questions.
TechCrunch AI
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
llama.cpp Releases
© Hugging Face BlogModels & Labscoding
IBM releases LLMON markup to fix prompt injection
IBM is tackling the brittleness of small language models by introducing LLMON, a structured markup language that explicitly separates instructions from data. By using reserved special tokens instead of ambiguous characters like JSON brackets, these new Granite 4.1 preview models drastically reduce indirect prompt injection attacks while maintaining standard performance. This approach forces the model to bind specific constraints to specific data payloads, eliminating the confusion that often leads to hallucinations or security breaches in long contexts.
Hugging Face Blog