# Thinking Machines vs Boundless

> Thinking Machines sells Tinker for custom post-training and serves its own Inkling models. Boundless serves open and closed models and offers no fine-tuning.

Canonical: https://www.subconscious.dev/compare/thinking-machines-vs-boundless · By The Subconscious Team · Updated October 29, 2026

## How they compare

Thinking Machines' Tinker exposes low-level training calls so teams write their own SFT or RL loops over LoRA adapters on models such as GLM-5.3, Kimi K2.6 and DeepSeek-V4.1-Flash, billed per million tokens across prefill, sample and train. Its beta serverless API serves only Inkling, at $1.00 in and $4.05 out per million tokens, and Inkling-Small, at $0.30 in and $1.20 out, both capped at 256K context there. Boundless offers no fine-tuning. It serves GLM-5.3 at $1.12 in and $3.52 out, DeepSeek V4.1 Flash at $0.20 in and $1.00 out and other open models, most with 1M context, plus resold GPT and Claude models, through an OpenAI and Anthropic compatible API.

Thinking Machines is the pick for research teams running their own post-training, including RL, and for trying the Inkling models. Boundless fits production serving of existing models, with per-key budgets on shared capacity or a reserved deployment its engineers tune and run against contract targets.

## What each one does

### Thinking Machines

Thinking Machines Lab is the San Francisco lab Mira Murati, formerly CTO of OpenAI, founded in February 2025, with OpenAI co-founder John Schulman as chief scientist. It raised about $2 billion at a $12 billion valuation in July 2025 in a round led by Andreessen Horowitz, and in March 2026 signed a multi-year Nvidia deal for one gigawatt of Vera Rubin capacity. In September 2026 it signed a $65M-a-year deal with Crusoe to serve Inkling, GLM 5.2 and 5.3 and its fine-tunes on dedicated B200s. Its main developer product is Tinker, an API for post-training open-weight models that launched in October 2025 and is now generally available. Tinker exposes four low-level calls, forward_backward, optim_step, sample and save_state, so teams write their own supervised or reinforcement learning loops while Thinking Machines runs the distributed GPU work. Training uses LoRA adapters rather than full weight updates.

### Boundless

Boundless launched in September 2026 as an inference partner for AI-native startups. It sells two things. The Model API serves pre-tuned models on shared capacity, billed per token from prepaid credits with a budget per API key. Tailored inference reserves capacity for one customer, and Boundless engineers pick the model on the customer’s own evals, tune precision, batching, caching, speculative decoding and parallelism, then run the deployment against uptime, throughput and latency targets written into the contract. A free two-week evaluation tests those targets on real traffic against the customer’s current provider first.

## Which is best, and when

### Choose Thinking Machines for

- Writing your own RL training loops
- LoRA fine-tunes of GLM-5.3 or Kimi K2.6
- Testing the Inkling models

### Choose Boundless for

- Production serving of open models
- Closed and open models on one key
- Contract latency and uptime targets

## At a glance

| Attribute | Thinking Machines | Boundless |
|---|---|---|
| Model access | Open weights | Open weights, plus closed GPT and Claude |
| Flagship models | Inkling, Inkling-Small | GLM-5.3, DeepSeek V4.1 Flash, Kimi K3 |
| Speed | - | 342 tok/s on DeepSeek V4.1 Flash 1017 (Vercel) |
| Price | Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out | $0.04–$2.30 in, $0.21–$11.40 out per 1M (open) |
| Customization | LoRA SFT and RL via Tinker | Serving-stack tuning, no fine-tuning |
| Deployment | Training API, beta serverless (Inkling only) | Serverless API, reserved tailored capacity |
| Long context | Inkling up to 1M (256K serverless); Tinker 32K–256K | 1M context on most models |

## FAQ

### What is the difference between Thinking Machines and Boundless?

Thinking Machines sells Tinker for custom post-training and serves its own Inkling models. Boundless serves open and closed models and offers no fine-tuning.

### When should I choose Thinking Machines over Boundless?

Writing your own RL training loops; LoRA fine-tunes of GLM-5.3 or Kimi K2.6; Testing the Inkling models.

### When should I choose Boundless over Thinking Machines?

Production serving of open models; Closed and open models on one key; Contract latency and uptime targets.

### Is Thinking Machines or Boundless cheaper?

Thinking Machines: Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out. Boundless: $0.04–$2.30 in, $0.21–$11.40 out per 1M (open). The cheaper choice depends on the model and workload.

### Which has more context, Thinking Machines or Boundless?

Thinking Machines: Inkling up to 1M (256K serverless); Tinker 32K–256K. Boundless: 1M context on most models.

## Running long-horizon agents?

If your agents run past 200K tokens, compare both against Subconscious: [Subconscious vs Thinking Machines](https://www.subconscious.dev/compare/subconscious-vs-thinking-machines.md), [Subconscious vs Boundless](https://www.subconscious.dev/compare/subconscious-vs-boundless.md).

Full profiles: [Thinking Machines](https://www.subconscious.dev/providers/thinking-machines.md), [Boundless](https://www.subconscious.dev/providers/boundless.md).