Introducing ASpec
vs

Thinking Machines vs TypeSafe AI

Two labs with ex-OpenAI founders taking different paths. Thinking Machines trains big open generative models. TypeSafe AI builds Jev, a decision model returning typed, calibrated answers in 70 to 500ms.

By The Subconscious Team · Updated

Thinking Machines vs TypeSafe AI: key differences

TypeSafe AI gives up text generation entirely. A developer defines the answer space with primitives like Choice, Score and a true-or-false type, and Jev returns a typed answer with calibrated probabilities. It scores every option in parallel in one pass, so calls finish in 70 to 500ms end to end, and TypeSafe says it runs roughly 40 to 200x faster than an LLM on decision-shaped queries. Thinking Machines stays generative. Inkling is a 975B MoE with 1M context and text, image and audio input, served in beta at $1.00 in and $4.05 out, and Tinker lets teams post-train open models with their own SFT or RL loops.

Jev fits inside software as a router, guardrail or smart if-statement, and since outputs always match the schema it cannot return a value outside the options. It is early access, text-only and has no customization listed. Thinking Machines covers what Jev cannot: writing, coding, reasoning over images and audio, and training a model to a team's own reward. Teams could train a classifier on Tinker, but they would need to host it elsewhere, since checkpoint sampling is scoped to testing. For fast routing decisions with a confidence score, Jev is the more direct tool.

What Thinking Machines and TypeSafe AI do

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.

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

Full Thinking Machines profile

TypeSafe AI

TypeSafe AI builds decision models instead of text generators. Founder Diogo Almeida co-invented RLHF and InstructGPT at OpenAI and later worked at Google Brain; his co-founders are Erik Gafni and Sasha Sheng. After two years in stealth the company launched on September 15, 2026 with a $40M seed round led by DCVC and released its first System One Model, Jev, in early access. The name nods to Kahneman's fast System 1 thinking, and the model is built for machines to call, not people to chat with. Weeks after launch, in October 2026, TypeSafe raised an $870M Series A led by Andreessen Horowitz at a $7.5B valuation. Jev 1.14 followed on October 14 at the same price, adding experimental image input and longer context behind opt-in flags.

Example models: Jev, jev-1.14

Full TypeSafe AI profile

Should you choose Thinking Machines or TypeSafe AI?

Thinking Machines

Choose Thinking Machines for

  • Generative tasks across text, image and audio input
  • Custom RL or SFT post-training
  • Long-context work up to 1M tokens

TypeSafe AI

Choose TypeSafe AI for

  • Routing and classification in 70 to 500ms
  • Guardrails with calibrated confidence
  • Schema-safe decisions inside agent harnesses

Thinking Machines vs TypeSafe AI at a glance

AttributeThinking MachinesTypeSafe AI
Model accessOpen weightsDecision models
Flagship modelsInkling, Inkling-SmallJev, jev-1.14
SpeedUnknown70–500ms per call
PricePer 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out$0.042 per 1M input tokens; output free
CustomizationLoRA SFT and RL via TinkerUnknown
DeploymentTraining API, beta serverless (Inkling only)Early-access API, plus Vercel AI Gateway, OpenRouter and LiteLLM
Long contextInkling up to 1M (256K serverless); Tinker 32K–256K64K per request, 32K per question

Frequently asked questions

What is the difference between Thinking Machines and TypeSafe AI?

Two labs with ex-OpenAI founders taking different paths. Thinking Machines trains big open generative models. TypeSafe AI builds Jev, a decision model returning typed, calibrated answers in 70 to 500ms.

When should I choose Thinking Machines over TypeSafe AI?

Generative tasks across text, image and audio input; Custom RL or SFT post-training; Long-context work up to 1M tokens.

When should I choose TypeSafe AI over Thinking Machines?

Routing and classification in 70 to 500ms; Guardrails with calibrated confidence; Schema-safe decisions inside agent harnesses.

Is Thinking Machines or TypeSafe AI cheaper?

Thinking Machines: Per 1M tokens by prefill, sample, train; Inkling $1.00 in, $4.05 out. TypeSafe AI: $0.042 per 1M input tokens; output free. The cheaper choice depends on the model and workload.

Which has more context, Thinking Machines or TypeSafe AI?

Thinking Machines: Inkling up to 1M (256K serverless); Tinker 32K–256K. TypeSafe AI: 64K per request, 32K per question.

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.