Models & Labs
Thinking Machines Lab Releases Inkling Model
WIRED AIhigh confidence
Why it matters
- →Inkling's open-weight design allows for greater customization and adaptation by researchers and startups.
- →The model's ability to handle multiple input types and perform advanced reasoning highlights its versatility.
- →Thinking Machines' approach challenges the dominance of closed models, promoting a more decentralized AI ecosystem.
Thinking Machines Lab, a startup founded by former OpenAI executives, has unveiled its first AI model, Inkling. This open-weight model, with 975 billion parameters, is capable of processing audio, video, and text inputs. Although it doesn't lead in benchmarks, it excels in reasoning and coding tasks. The model's open-source nature allows for customization, supporting the company's goal of decentralizing AI development. This move positions Thinking Machines as a formidable player in the competitive AI field.
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 · Same story
- 2Thinking Machines unveils interactive AI model
TechCrunch AI · May 12, 2026 · Same story
- 3TML Unveils Real-Time AI Interaction Models
The Rundown AI · May 12, 2026 · Related
- 4Thinking Machines Unveils New Interaction Models
The AI Daily Brief · May 13, 2026 · Related
- 5Mira Murati's Startup Aims for Human-Centric AI
WIRED AI · May 15, 2026 · Background
- 6Thinking Machines Labs Unveils New Interaction Models
Matt Wolfe · May 16, 2026 · Related
- 7Microsoft Unveils MAI-Thinking-1 AI Model
The Verge AI · June 2, 2026 · Related
- 8Open Models Propel AI Research at ICML 2026
NVIDIA Blog · July 6, 2026 · Background
What happened next
- 9Debate Over Open-Weight AI Models Intensifies
TechCrunch AI · July 20, 2026 · Background
- 10Chinese AI Models Challenge Silicon Valley's Dominance
WIRED AI · July 22, 2026 · Background
- 11Chinese Labs Lead in Open Model Scale
Hugging Face Blog · August 14, 2026 · Background
- 12Liquid AI opens d1 decision models for edge inference
Hugging Face Blog · October 7, 2026 · Related
- 13MIT shows AI rewriting model weights without training
The Rundown AI · November 3, 2026 · Background
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 - Jul 15 · WIRED AI
Thinking Machines Lab Releases Inkling Model (This article) - Jul 20 · Fireship
Thinking Machines releases Inkling model
More from WIRED AI
© WIRED AIMarket & Regulationbusiness
Nvidia Names Interim CEO Colette Kress
Nvidia’s board has installed CFO Colette Kress as interim CEO following Jensen Huang’s resignation amid a scandal involving alleged Chinese intelligence ties. This is not just a personnel shuffle; it marks the first time Nvidia operates without its visionary founder at the helm, shifting power to the executive who knows the company's financials and supply chain best. While Kress lacks Huang’s engineering pedigree, her deep familiarity with CUDA’s economics and investor relations provides immediate stability during an active federal investigation. The market now faces a leadership vacuum where technical direction must be managed by a team rather than a single charismatic figurehead.
WIRED AI
© WIRED AIMarket & Regulationother
Spirit Airlines Union Secures AI Data Opt-Out
A landmark settlement between Google and the Association for Flight Attendants establishes a new precedent for worker data privacy in the age of AI training. By mandating third-party de-identification via Tonic AI before selling Spirit Airlines' massive dataset, the deal protects billions of employee records from being ingested without consent. This is not just about one airline's liquidation; it creates a replicable legal framework that unions can demand in future corporate transactions. It signals that worker data rights are becoming a negotiable asset class, forcing tech giants to build privacy safeguards into their data acquisition pipelines.
WIRED AI
© WIRED AIGeneral AIother
Experts pivot AI safety from extinction to cyber and bio threats
The AI safety narrative has shifted from speculative extinction scenarios to tangible infrastructure risks. A WIRED panel of experts identifies nuclear integration errors, cyberattacks on critical systems, and biological weaponization as the immediate dangers. This marks a move away from abstract 'runaway intelligence' fears toward verifiable industrial hazards like flawed defense workflows and unregulated bio-threats. The consensus suggests that current lab-driven safety thresholds are insufficient for these concrete threats.
WIRED 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
© 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