Thinking Machines Lab
AI lab building frontier models and fine-tuning tools like Tinker, founded by Mira Murati.
On the record
Checkpoint: Resolution of stalled funding round and release of Inkling-Small model weights
The call
Thinking Machines Lab has cash and talent but is watching both leak out while bigger, cheaper rivals copy its playbook.
Thinking Machines Lab helps companies customize AI models for their specific needs through a service called Tinker. The problem is that big tech companies like Meta and OpenAI can do this themselves now, and cheaper DIY tools exist that do similar work for a quarter of the price. What might save the company is a special promise to customers in regulated industries: they say they won't keep your data and you can move your AI models elsewhere, but this advantage has never been tested against well-funded rivals like Fireworks AI, which just raised even more money with the exact same idea.
Why
01
AI pressure: Fireworks AI. Fireworks AI raised $1.5B pitching the same managed fine-tuning and inference thesis Tinker sells, at a much larger valuation. Could AI do the core job? Yes, likely.
02
Bigger players: Meta and OpenAI. Meta and OpenAI are pulling Thinking Machines co-founders and staff away, while OpenAI ships its own native fine-tuning API. Could a giant copy it? Yes, easily.
03
What protects it. A no-retention data policy and checkpoint portability aimed at regulated customers, still unproven against richer rivals.
What would change this call
Whether Thinking Machines closes its stalled ~$50B funding round in 2026. Would show if investors still back the moat despite talent losses.
Further co-founder or senior researcher departures to Meta, OpenAI, or xAI. Would confirm talent attrition is accelerating, not stabilizing.
Whether Fireworks AI or Together AI publicly report enterprise wins in fine-tuned inference. Would show rivals converting Tinker's target customers to their platforms.
Further co-founder or senior researcher departures to Meta, OpenAI, or xAI. Would confirm talent attrition is accelerating, not stabilizing.
Whether Fireworks AI or Together AI publicly report enterprise wins in fine-tuned inference. Would show rivals converting Tinker's target customers to their platforms.
Key numbers
- $2B in capital raised
- Fireworks AI raised $1.5B at $17.5B valuation
- Tinker priced 4-5x above DIY fine-tuning costs
Named risks
- No disclosed revenue while committing to gigawatt-scale Nvidia compute costs.
- Continued co-founder and talent departures to Meta, OpenAI, and xAI.
- Tinker commoditization risk as DIY fine-tuning can be 4-5x cheaper at scale.
- Stalled $50B funding round raises questions about investor appetite.
- Reported internal governance issue: co-founder Barret Zoph's termination cited for unethical conduct.
Rivals, as read in July 2026
| Rival | Direction | Evidence |
|---|---|---|
| Meta | losing to | Meta reportedly offered $1B to acquire the company, then hired 5 founders individually by April 2026 |
| OpenAI | losing to | Co-founders Barret Zoph and Luke Metz returned to OpenAI, January 2026 |
| Fireworks AI | contesting | Raised $1.5B at $17.5B in July 2026 targeting same fine-tuned inference niche as Tinker |
| DeepSeek, Qwen, Kimi (Chinese open-weight labs) | losing to | Lead several open-weight benchmarks as of July 2026, prompting Inkling as a Western counter |
| xAI | losing to | One founding member departed for xAI, per April 2026 reporting |
Questions people ask
- Is Thinking Machines Lab profitable in 2026?
- No revenue figures are public. TechCrunch reported on 2026-07-15 that revenue has not been a priority for Thinking Machines Lab, which is instead focused on shipping products like Tinker and Inkling while burning through its $2B war chest.
- Is Thinking Machines Lab losing talent to Meta and OpenAI?
- Yes. By April 2026, five original founders had left for Meta including Andrew Tulloch on a reported $1.5B package, three returned to OpenAI (including co-founders Barret Zoph and Luke Metz in January 2026), and one joined xAI.
- Did Mark Zuckerberg try to buy Thinking Machines Lab?
- Reports say Meta CEO Mark Zuckerberg offered roughly $1 billion to acquire Thinking Machines Lab outright; Mira Murati rejected the offer, after which Meta shifted to individually hiring several founding members, per an April 21, 2026 report.
- Who is Thinking Machines Lab's biggest competitor?
- Fireworks AI is the closest direct rival, raising $1.5B at a $17.5B valuation in July 2026 targeting the same specialized fine-tuned inference niche as Tinker. Mistral, Together AI, Cohere, and Chinese labs DeepSeek, Qwen, and Kimi also compete in open-weight AI.
- What is Tinker, Thinking Machines Lab's main product?
- Tinker is a managed API for fine-tuning open-weight models like Llama, Qwen3, DeepSeek-V3.1, and Kimi, launched October 1, 2025. It reached general availability in December 2025 and charges usage-based per-million-token pricing across three meters.
- What is Inkling, Thinking Machines Lab's AI model?
- Inkling is Thinking Machines Lab's first in-house model, launched July 15, 2026, as an open-weight mixture-of-experts model with 975B total parameters (41B active), multimodal input, and 1M-token context, released under Apache 2.0 and trained on Nvidia GB300 NVL72 systems.
Timeline
- 2026-07-15Launches Inkling, first in-house open-weight model, under Apache 2.0 license
- 2026-03Nvidia makes strategic investment, commits 1 gigawatt of Vera Rubin compute capacity
- 2026-01-14Co-founders Barret Zoph and Luke Metz leave for OpenAI
- 2025-11Reportedly seeks new funding round at $50 billion valuation
- 2025-10-01Launches Tinker, fine-tuning API for open-weight models
- 2025-06-22Closes $2B seed round at $10 to 12 billion valuation led by Andreessen Horowitz
- 2025-02-20Mira Murati launches company with about 30 researchers from OpenAI, Meta, Mistral
Revision history
20 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources