Together AI brings Thinking Machines Lab’s new model Inkling on day 0

2026-07-15 · Source: Together AI | The AI Native Cloud - Together.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure · Depth: Advanced, medium

Summary

Thinking Machines Lab has released Inkling, a new multimodal mixture-of-experts model, now available on Together AI's production inference platform as of July 15, 2026. Designed for token-efficient reasoning and native multimodal understanding, Inkling accepts text, image, and audio inputs to produce text outputs. The model features a 975B total parameter count with 40B active parameters per token and a 1M token context window. Its architecture incorporates query-conditioned relative attention, short causal convolutions, and a shared expert sink Mixture-of-Experts design. Preliminary evaluations show strong performance across scientific reasoning, mathematics, coding, agentic workflows, forecasting, and calibrated prediction tasks. Together AI provides optimized inference, including a FlashAttention-4-based attention kernel, enabling developers to access Inkling with day-zero availability and controllable reasoning effort via a unified API.

Key takeaway

For AI Engineers building advanced multimodal applications, Inkling offers a compelling solution for integrating text, image, and audio inputs through a single model. You can utilize its controllable reasoning effort via API to optimize for latency, token usage, and reasoning depth per request, eliminating the need for complex infrastructure management. This simplifies development and deployment, allowing you to focus on application logic rather than managing separate preprocessing pipelines or GPU resources.

Key insights

Inkling unifies multimodal inputs and offers controllable reasoning via a novel MoE architecture for broad task versatility.

Principles

• Unified decoder architectures can process diverse modalities.
• Inference effort can be dynamically adjusted for efficiency.
• Query-conditioned relative attention improves token order representation.

Method

Inkling transforms image patches and quantized audio into embeddings, integrating them directly into the decoder's input sequence for joint reasoning.

In practice

• Adjust "reasoning_effort" via API for cost/speed tuning.
• Utilize single API endpoint for all text, image, audio inputs.
• Leverage 1M token context window for complex multimodal tasks.

Topics

• Inkling
• Multimodal AI
• Mixture-of-Experts
• AI Inference
• Together AI
• Reasoning Models

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• Thinking Machines Lab Drops Its First Model — Thinking Machines Lab launched Inkling, a large open-source multimodal model, challenging established AI players with a vision for decentralized AI.
• not much happened today — Inkling's launch emphasizes an open-weight, customizable multimodal foundation model designed for practical application and ecosystem integration over pure benchmark leadership.
• [AINews] Thinky's Inkling: 975B-A41B multimodal, new best American Apache 2.0 open model (with Inkling-Small, 276B-A12B) — Inkling is an Apache 2.0 multimodal MoE model, prioritizing customization and efficiency over raw benchmark supremacy.
• Inkling model from Thinking Machines Lab now on Databricks — Inkling, an open-weights coding model, is now on Databricks, offering enterprises customizable, secure, and cost-effective AI for coding workflows.
• The AI That Can Re-Write Its Own Brain: Why Inkling is the New Frontier for Open Weights — Inkling enables autonomous AI development and dynamic intelligence scaling through an open-weights, self-modifying architecture.
• Complete Guide to Thinking Machines Inkling — Inkling is an open-weights multimodal MoE model prioritizing customization and domain-specific fine-tuning over benchmark supremacy.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Together AI | The AI Native Cloud - Together.ai.