AI agents · OpenClaw · self-hosting · automation

Quick Answer

Thinking Machines vs SSI vs Anthropic: Founders (Jul 2026)

Published:

Thinking Machines vs SSI vs Anthropic: Founders (Jul 2026)

On July 15, 2026, Mira Murati’s Thinking Machines Lab shipped Inkling — its first in-house open-weight model. That completes an ex-OpenAI leadership trilogy that has been reshaping the AI market:

  • Anthropic (Dario & Daniela Amodei, 2021) — the first ex-OpenAI spinout, now a mature frontier lab.
  • SSI (Safe Superintelligence Inc.) (Ilya Sutskever, June 2024) — the pure “straight shot to safe superintelligence” bet.
  • Thinking Machines Lab (Mira Murati, February 2025) — the newest, betting on customization + open weights.

Three founders, three fundamentally different strategies. Here’s how they actually differ in practice as of July 2026.

Last verified: July 19, 2026

The Founders and the Founding Bets

LabFounder(s)FoundedCore Bet
AnthropicDario Amodei (CEO), Daniela Amodei (President)April 2021Safety-first frontier models shipped as commercial products
SSIIlya Sutskever (with Daniel Gross, Daniel Levy)June 2024One long research push toward safe superintelligence; no interim products
Thinking MachinesMira Murati (CEO), ex-OpenAI research teamFebruary 2025Customization-over-one-size-fits-all; open-weight base + Tinker platform

Head-to-Head — Products, Revenue, Valuation

DimensionAnthropicSSIThinking Machines
Shipped productsClaude (Sonnet 5, Fable 5, Mythos 5, Opus 4.8), Claude Code, Claude CoworkNone (by design)Inkling (open-weight model, July 15, 2026), Tinker (fine-tuning platform)
Est. annualized revenue (mid-2026)$6-10B+ ARR (reported ranges)$0 (no shipped product)Not disclosed; Tinker generates some usage revenue
Latest funding valuation~$61B (Series H, early 2026); rumored $100B+ round Q2/Q3 2026$32B (April 2025 round; no public update since)$12B post-money (2025 seed); rumored $50-60B follow-on stalled early 2026
Total capital raised (public)~$18B+ (Amazon, Google, others)~$1B (April 2025; further undisclosed)$2B seed (2025); undisclosed since
Employee count (approx)~2,000+ (2026)~40-60 (small by design)~200-300
Public commercial APIYes (Claude API + AWS Bedrock + Google Vertex)NoInkling on Hugging Face; Tinker as service
Open-weight releasesNone (closed)NoneInkling (Apache 2.0, July 15, 2026)
Fine-tuning platform for customersClaude fine-tuning (limited)NoneTinker (the core commercial hook)

The Strategic Bets Compared

Anthropic — Safety-first at frontier scale

The bet: Build the safest frontier models and sell them as commercial products. Use commercial revenue to fund research on scalable alignment.

How it’s playing out:

  • Claude Sonnet 5 (June 30, 2026) is the default in Claude Code and widely deployed in enterprise coding stacks.
  • Claude Fable 5 competes head-to-head with GPT-5.6 Sol at the frontier.
  • Claude Cowork (July 7, 2026) extends conversations across web / mobile / desktop.
  • Enterprise product line (Claude for Enterprise, device verification for Team/Enterprise plans as of July 16, 2026) is deepening.
  • Reported ARR of $6-10B+ is the highest of the three.

Where it might break: Frontier training costs. Anthropic’s 2026 spend is likely $10B+ on compute; requires ongoing $10B+/year fundraising to maintain the pace. If GPT-5.6 / Gemini 3.5 pull materially ahead on capability, revenue erodes.

SSI — The straight shot

The bet: Skip commercial products entirely. One long research push toward safe superintelligence. Ship nothing until “the plan is ready.”

How it’s playing out:

  • No public model.
  • No API.
  • No published benchmarks.
  • Small team by design (~40-60 employees, mostly research).
  • $32B valuation (April 2025) is entirely a bet on Ilya + team + strategy.

Where it might break: Two ways:

  1. Their approach doesn’t produce a demonstrable superintelligence — investors demand a product pivot, which they refuse.
  2. Frontier labs (Anthropic, OpenAI, Google DeepMind) reach superintelligence first through iterative shipping, making SSI’s approach look wrong-headed.

Where it might win: If iterative shipping proves fundamentally unsafe or capped, and only a from-scratch safety-first architecture works, SSI wins the whole board.

Thinking Machines — Customization + open weights

The bet: Frontier isn’t where the durable value is. Customization to specific enterprise domains is. Ship an open-weight base model (Inkling) + a fine-tuning platform (Tinker) so enterprises can build their own models on top.

How it’s playing out:

  • Inkling (July 15, 2026) — 975B / 41B-active MoE, Apache 2.0.
  • Tinker platform — fine-tuning-as-a-service on Inkling and other open-weight bases.
  • Nvidia investment + partnership signals infrastructure alignment.
  • $12B seed valuation reflected excitement; the follow-on round stalling reflects skepticism.

Where it might break: If enterprises don’t fine-tune at scale (they mostly use closed frontier models today), the “customization is the durable market” thesis doesn’t play out. Inkling by itself is not a frontier-topping model — Kimi K3 (July 17, 2026) outscores it — so the vehicle only works if Tinker turns the wheel.

Where it might win: If US-based Apache 2.0 open weights + integrated fine-tuning becomes the enterprise standard for regulated / customized deployments, Thinking Machines becomes the AWS-of-fine-tuning. The market for enterprise customization is plausibly $20B+/year by 2028.

Sub-Questions People Are Asking

Why did all three founders leave OpenAI?

Publicly-known reasons vary:

  • Amodeis (2021): disagreement over safety-first direction as OpenAI pivoted to commercial GPT-3 monetization.
  • Sutskever (2024): post-board-drama exit; publicly articulated as safety focus.
  • Murati (2024): unclear public reasons; formally left CTO role in September 2024, founded Thinking Machines in February 2025.

The common thread is “we can do this differently than the current OpenAI model” — but each interprets “differently” in a different direction.

Which lab attracts the strongest talent?

Anthropic has the deepest bench of published alignment researchers. SSI has Ilya + Daniel Gross + Daniel Levy (small, elite). Thinking Machines has a strong cohort of ex-OpenAI research leads (John Schulman was reportedly recruited but eventually returned to Anthropic; other senior researchers stayed at Thinking Machines).

Talent is fluid — but Anthropic dominates on total senior AI researcher headcount today.

Are these labs competitive with the closed frontier majors?

  • Anthropic: yes, directly. Claude Fable 5 / Opus 4.8 are frontier-competitive with GPT-5.6 / Gemini 3.5.
  • SSI: unknown — no shipped products.
  • Thinking Machines: not on raw capability. Inkling trails Kimi K3 and Claude Fable 5 on benchmarks. But that’s not the point of Inkling — it’s a customization base.

Which one is most fundable in a downturn?

In a real AI-spending pullback, Anthropic is safest — has revenue to fund itself partially. SSI is riskiest — no revenue, no product, entirely dependent on continued investor confidence in Ilya. Thinking Machines is middle-risk — some Tinker revenue, but the enterprise-customization thesis is unproven.

Is Ilya’s SSI actually going to ship anything?

The stated plan is one long training run to safe superintelligence with no interim products. As of July 2026, that plan is still the plan (no product pivot announced). Investors either believe in it or don’t — there’s no evidence to check either way.

Which one does the enterprise buyer default to?

Anthropic. Claude API + AWS Bedrock Claude access + Claude Code + Claude Cowork = a full enterprise stack. Thinking Machines is worth evaluating for customized workflows. SSI has nothing to buy.

What’s the read on Thinking Machines’ stalled follow-on round?

Reports (late 2025 / early 2026) suggest Thinking Machines was in discussions for a follow-on at $50-60B pre-money that stalled around the start of 2026 without closing. Public interpretations vary:

  • Bull case: the company chose to wait for Inkling launch to price the round.
  • Bear case: investors balked at the valuation without shipped products.

Inkling shipping July 15 is likely the trigger for a fresh fundraise attempt in H2 2026.

The Real Decision Framework

For enterprise buyers:

  • Default choice: Anthropic (Claude Sonnet 5 / Fable 5) — production-ready, mature enterprise features.
  • Add-on for customization: Thinking Machines (Inkling + Tinker) — for US-procurement-restricted fine-tuning workloads.
  • Ignore for procurement decisions: SSI — no product to buy or integrate.

For investors:

  • Highest cash-flow safety: Anthropic (revenue-scaling frontier lab).
  • Highest optionality: SSI (bet on Ilya + strategy).
  • Middle-ground bet on emerging customization market: Thinking Machines.

For talent:

  • Anthropic if you want to ship at scale to a real user base.
  • SSI if you want small-team, long-horizon research with elite peers.
  • Thinking Machines if you want to build the customization / open-weight platform layer.

For AI market watchers:

  • Anthropic is the frontier-competitive incumbent to compare against OpenAI, Google, xAI.
  • SSI is the wild card — could redefine the market or fade.
  • Thinking Machines is the platform bet — could become the AWS of open-weight enterprise AI, or fade if customization doesn’t emerge as a durable market.

Bottom Line

Three founders left OpenAI. Three fundamentally different bets. As of July 2026:

  • Anthropic ships and monetizes at scale. The enterprise default. Revenue is real.
  • SSI ships nothing and won’t for years. The pure strategic option. Unfalsifiable until it succeeds or the money runs out.
  • Thinking Machines just shipped Inkling — the first US-origin frontier-class open-weight model. Success depends on whether enterprise customization emerges as the durable market Murati is betting it will.

For enterprises today: use Anthropic. Evaluate Thinking Machines for customization workloads. Watch SSI without operational dependence.

For the market long-term: the three bets are structurally different enough that all three could succeed in non-overlapping niches — or two could fail if the market consolidates around one strategy. The next 18 months will start to answer that.

Sources