Anthropic vs Thinking Machines

Anthropic sells finished Claude models for agentic coding. Thinking Machines sells the tools to post-train an open model yourself, plus its own Inkling weights.

By The Subconscious Team · Updated September 30, 2026

Anthropic vs Thinking Machines: key differences

Anthropic offers the Claude lineup, from Fable 5.1 at $10 in and $50 out down to Haiku 4.5 at $1 in and $5 out, with a 1M window on the top three tiers and no surcharge past 200K. It runs on its own API, Bedrock, Vertex AI and Microsoft Foundry, and Claude Code made it a default inside many engineering teams. Claude is sold as finished closed models, not as weights to customize. Thinking Machines comes at the problem from the weights up. Tinker lets teams run LoRA-based SFT or RL on open models like Kimi K2.6, GLM-5.3, DeepSeek-V3.1 and Inkling, writing their own loop while the lab handles distributed GPU work. Billing is per million tokens across prefill, sample and train.

For shipping an agent today, Anthropic is the stronger product: top-tier SWE-bench Pro results, cheap $0.25 cache reads on Fable 5.1, and enterprise procurement on every major cloud. Thinking Machines' serving is thin, with a beta serverless API for Inkling only and a checkpoint endpoint scoped to low internal traffic. Its case rests on ownership. Inkling is a 975B-parameter Apache 2.0 MoE with 41B active, 1M context and native audio input, priced at $1.00 in and $4.05 out. Teams that need a narrow task model they control, or want RL on their own reward signal, get something Claude cannot offer.

What Anthropic and Thinking Machines do

Anthropic

Anthropic sells the Claude family of closed models through its own API, Amazon Bedrock, Google Vertex AI and Microsoft Foundry. The public lineup today runs from Claude Fable 5.1 at the top, released September 1, 2026, through the Opus and Sonnet tiers down to Haiku 4.5. List prices span a tenfold range, from $10 in and $50 out on Fable to $1 in and $5 out on Haiku. The top three tiers include a 1M token context window at standard pricing with no surcharge past 200K.

Example models: Claude Fable 5.1, Claude Haiku 4.5

Full Anthropic profile

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. 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.

Example models: Inkling, Inkling-Small, Qwen3.5, Kimi K2.6

Full Thinking Machines profile

Should you choose Anthropic or Thinking Machines?

Anthropic

Choose Anthropic for

Agentic coding inside Claude Code
Enterprise procurement through Bedrock, Vertex AI or Foundry
Long agent loops using cheap Fable 5.1 cache reads

Thinking Machines

Choose Thinking Machines for

Training a task-specialized model on an open base
RL on a custom reward signal with full loop control
Open 1M-context models with native audio input

Anthropic vs Thinking Machines at a glance

Attribute	Anthropic	Thinking Machines
Model access	Closed	Open weights
Flagship models	Claude Fable 5.1, Opus, Sonnet, Haiku 4.5	Inkling, Inkling-Small
Speed	Fable is the slowest tier	Unknown
Price	$1–$10 in, $5–$50 out per 1M	Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out
Customization	N/A	LoRA SFT and RL via Tinker
Deployment	API, Bedrock, Vertex AI, Microsoft Foundry	Training API, beta serverless (Inkling only)
Long context	1M, no surcharge past 200K	Inkling up to 1M; Tinker 32K–256K

Frequently asked questions

What is the difference between Anthropic and Thinking Machines?

Anthropic sells finished Claude models for agentic coding. Thinking Machines sells the tools to post-train an open model yourself, plus its own Inkling weights.

When should I choose Anthropic over Thinking Machines?

Agentic coding inside Claude Code; Enterprise procurement through Bedrock, Vertex AI or Foundry; Long agent loops using cheap Fable 5.1 cache reads.

When should I choose Thinking Machines over Anthropic?

Training a task-specialized model on an open base; RL on a custom reward signal with full loop control; Open 1M-context models with native audio input.

Is Anthropic or Thinking Machines cheaper?

Anthropic: $1–$10 in, $5–$50 out per 1M. Thinking Machines: Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out. The cheaper choice depends on the model and workload.

Which has more context, Anthropic or Thinking Machines?

Anthropic: 1M, no surcharge past 200K. Thinking Machines: Inkling up to 1M; Tinker 32K–256K.

Related comparisons

Running long-horizon agents?

If your agents run past 200K tokens, compare both against Subconscious. Our inference stack treats a long-horizon trace as the primary workload, so speed, cost, and accuracy hold up deep into the trace.

Subconscious vs Anthropic Subconscious vs Thinking Machines