Inkling AI: Mira Murati's Lab Launches Top Western Open Model
Mira Murati's Thinking Machines Lab released Inkling, an open-source AI model with strong agentic tool use, outperforming Western rivals on MCP Atlas and SWE-Bench. However, it trails Chinese models on raw coding benchmarks and cost efficiency.
Quick Take
Inkling achieves 74.1% on MCP Atlas and 77.6% on SWE-Bench, beating NVIDIA's Nemotron.
The model has 975B total parameters, 41B active, and supports text, image, and audio.
Priced at $1 per million input tokens on OpenRouter, it's suitable for agentic workflows.
Despite strengths, Inkling underperforms Chinese models in raw coding performance per dollar.
Market Impact Analysis
NeutralThe article is about an AI model release with no direct implications for cryptocurrency markets.
Speculation Analysis
Key Takeaways
- Inkling achieves 74.1% on MCP Atlas and 77.6% on SWE-Bench Verified, handily beating NVIDIA's Nemotron for agentic tool use.
- The mixture-of-experts model packs 975B total parameters with 41B active at inference, supporting text, image, and audio inputs.
- Priced at $1 per million input tokens on OpenRouter, it targets developers building agentic workflows without breaking the bank.
- Raw coding benchmarks still favor Chinese models from Alibaba and Moonshot AI, limiting Inkling's edge in cost-sensitive deployments.
What Happened
Mira Murati's stealth startup, Thinking Machines Lab, released its first model on July 15. Inkling is a 975-billion-parameter open-source AI built from the ground up over two years. The model went live on OpenRouter and Hugging Face under the permissive Apache 2.0 license, giving developers immediate access to its weights with no regional restrictions. It is a mixture-of-experts architecture: 41 billion parameters are active during inference, while a 1-million-token context window swallows large prompts. The launch positions Inkling as the strongest open-source model trained by a Western lab this year, directly challenging Chinese dominance on public leaderboards.
The Numbers
Inkling was pretrained on 45 trillion tokens and handles text, images, and audio. Its benchmark scores tell the story. On MCP Atlas, which measures agentic tool use reliability, it scored 74.1%—nearly 30 points above NVIDIA's Nemotron 3 Ultra. On SWE-Bench Verified, a hard test of real-world GitHub issue resolution, it hit 77.6% versus Nemotron's 70.7%. These numbers make it a top pick for agentic workflows. Pricing on OpenRouter sits at $1 per million input tokens and $4.05 per million output tokens, slotting it into the mid-tier cost bracket for large models.
Why It Happened
Murati left OpenAI in September 2024 and spent two years building a lab that could ship a competitively open model. Western labs have ceded ground to Chinese teams like Alibaba's Qwen and Moonshot AI's Kimi, which dominate raw coding benchmarks. Inkling is a direct answer: a no-strings-attached, Apache 2.0 release designed to win developer mindshare in the agentic tooling space. The focus on MCP Atlas performance signals a bet that the next wave of AI utility will hinge on how reliably models can use external tools, not just generate code.
Broader Impact
Inkling lowers the barrier for building agentic applications without proprietary lock-in. Its permissive license and competitive agentic scores could accelerate adoption in decentralized or crypto-native projects that prize composability. However, Chinese models still offer better raw coding per dollar, keeping pressure on Thinking Machines Lab to iterate quickly or risk being undercut on cost. The release nonetheless marks a notable Western pushback in the open-source AI race that had been tilting East.
What to Watch Next
- Developer uptake via OpenRouter and Hugging Face. If agentic workflow builders switch to Inkling, it validates Murati's bet on tool-use benchmarks.
- Chinese model releases. Any jump in MCP Atlas scores from Qwen, GLM, or Kimi could blunt Inkling's advantage within weeks.
- Pricing adjustments. OpenRouter costs are already competitive, but a price war with DeepSeek-style ultra-low-cost models could reshape the landscape.
This article is for informational purposes only and does not constitute financial advice.
Always late to trends?
Join for the latest news, insights & more.
Disclaimer: Bytewit is an independent media outlet that delivers news, research, and data.
© 2026 Bytewit. All Rights Reserved. This article is for informational purposes only.
