Is Goodfire profitable in 2026?

No revenue or profitability figures have been disclosed. Goodfire raised a $150M Series B at a $1.25B valuation on 2026-02-05, but no ARR, customer counts, or profit metrics have been made public as of July 2026.

Could AI labs replace Goodfire's interpretability product?

Yes, partly. Anthropic, an investor in Goodfire, published overlapping 'J-lens/J-space' interpretability research on Claude Opus 4.6 in July 2026, showing frontier labs can build similar tools in-house rather than buying Goodfire's Ember API.

Goodfire is privately held and venture-backed. Its Series B closed 2026-02-05 with $150M led by B Capital at a $1.25B valuation, with investors including Anthropic, Menlo Ventures, Lightspeed, Salesforce Ventures, and Eric Schmidt.

Is Goodfire's biggest competitor Anthropic?

Anthropic is unusual: it is both an investor (its first-ever startup investment, per an April 2025 report) and a direct competitor, having shipped its own interpretability research (J-lens) on Claude Opus 4.6 in July 2026 that overlaps Goodfire's Ember product.

What does Goodfire's product Ember actually do?

Ember is Goodfire's hosted mechanistic-interpretability API and SDK, launched December 2024, that exposes internal 'features' of models like Llama 3.1 8B and Llama 3.3 70B so users can search, observe, and steer model behavior at inference time.

Has Goodfire produced any scientific results outside AI models?

Yes. Its partner Prima Mente used Goodfire's interpretability methods to identify a novel class of Alzheimer's biomarkers, announced around Goodfire's Series B on 2026-02-05, described as the first major natural-science finding from reverse-engineering a foundation model.

United States · United States

Goodfire

Goodfire builds tools that decode what is happening inside AI models, sold via API and science partnerships.

Losing its protection

Assessed 29 July 2026 · Evidence as of 29 July 2026

The structural protection is eroding.

On the recordSince 29 July 2026 · Checkable from H2 2026 · Every call on the record Checkpoint: Anthropic's next interpretability release: whether it becomes a hosted API for outside customers or stays internal

The call

Goodfire sits between frontier labs that could build its tools in-house and an investor, Anthropic, that already shipped a rival version.

Goodfire sells software that helps researchers and companies look inside large language models to see how they work and catch problems. The hard part is that Anthropic, which invested in Goodfire, has already released its own competing tool called J-lens and can keep improving it without paying anyone. The company's only real advantage right now is early partnerships with research labs and the talent of its founding team, but those advantages could disappear if Anthropic keeps building better tools for free or if Goodfire's researchers leave to work at bigger AI companies. Everything changes if Goodfire can prove people will actually pay money for what it sells before Anthropic and other labs decide interpretability is basic enough to build themselves.

Why

AI pressure: Anthropic (J-lens/J-space interpretability tool). Anthropic, also Goodfire's investor, shipped its own interpretability tool in July 2026, narrowing Goodfire's edge on Claude specifically. Could AI do the core job? Only partly.

Bigger players: OpenAI, Google DeepMind. Both labs run in-house interpretability research and could fold it into their own products without needing Goodfire. Could a giant copy it? Yes, easily.

What protects it. A neutral, cross-model interpretability layer plus science partnerships (Arc Institute, Mayo Clinic, Prima Mente) that no single lab owns.

What would change this call

Anthropic offers J-lens or a successor as a hosted API for third-party Claude customers. Would narrow Goodfire's enterprise wedge to only non-Claude models.
Goodfire discloses ARR figures or a Series C valuation mark within 12-18 months. A flat or absent raise would signal weakening investor confidence.
A senior Goodfire researcher departs for Anthropic, OpenAI, or Google DeepMind. Would show the talent seal is leaking to frontier labs.

Key numbers

$150M Series B at $1.25B valuation (2026-02-05)
$50M Series A, Anthropic's first startup investment (2025-04-17)
Science partners: Arc Institute, Mayo Clinic, Prima Mente (2026-02-05)

Named risks

Anthropic, a direct investor, is also a competitor shipping overlapping interpretability research in-house. Severity: high
No disclosed revenue, ARR, or customer counts as of July 2026. Severity: high
Small senior research team vulnerable to poaching by frontier labs without any acquisition. Severity: medium
Unclear product transition between Ember and newer Silico branding could disrupt customer contracts. Severity: medium
Rapid scaling post Series B creates execution risk moving from small team to enterprise delivery. Severity: low

Rivals, as read in July 2026

Rival	Direction	Evidence
Anthropic	contesting	Anthropic, an investor since Apr 2025, shipped overlapping J-lens interpretability work on Claude Opus 4.6 in July 2026
Tilde	losing to	Tilde shipped its Sieve SAE-based control pipeline before Goodfire's Ember launched in Dec 2024, per an Aug 2025 analyst report
OpenAI	contesting	Maintains internal interpretability teams that could substitute for Goodfire's third-party tools, per Aug 2025 analysis
Google DeepMind	contesting	Source of Goodfire's own Chief Scientist Tom McGrath, who led its interpretability team before co-founding Goodfire
EleutherAI and InterpretML	contesting	Free open-source interpretability tooling pressures Goodfire's paid Ember product, per Aug 2025 report

Questions people ask

Is Goodfire profitable in 2026?: No revenue or profitability figures have been disclosed. Goodfire raised a $150M Series B at a $1.25B valuation on 2026-02-05, but no ARR, customer counts, or profit metrics have been made public as of July 2026.
Could AI labs replace Goodfire's interpretability product?: Yes, partly. Anthropic, an investor in Goodfire, published overlapping 'J-lens/J-space' interpretability research on Claude Opus 4.6 in July 2026, showing frontier labs can build similar tools in-house rather than buying Goodfire's Ember API.
Who owns Goodfire?: Goodfire is privately held and venture-backed. Its Series B closed 2026-02-05 with $150M led by B Capital at a $1.25B valuation, with investors including Anthropic, Menlo Ventures, Lightspeed, Salesforce Ventures, and Eric Schmidt.
Is Goodfire's biggest competitor Anthropic?: Anthropic is unusual: it is both an investor (its first-ever startup investment, per an April 2025 report) and a direct competitor, having shipped its own interpretability research (J-lens) on Claude Opus 4.6 in July 2026 that overlaps Goodfire's Ember product.
What does Goodfire's product Ember actually do?: Ember is Goodfire's hosted mechanistic-interpretability API and SDK, launched December 2024, that exposes internal 'features' of models like Llama 3.1 8B and Llama 3.3 70B so users can search, observe, and steer model behavior at inference time.
Has Goodfire produced any scientific results outside AI models?: Yes. Its partner Prima Mente used Goodfire's interpretability methods to identify a novel class of Alzheimer's biomarkers, announced around Goodfire's Series B on 2026-02-05, described as the first major natural-science finding from reverse-engineering a foundation model.

Timeline

2026-07-09Anthropic publishes J-lens/J-space interpretability research on Claude Opus 4.6, overlapping Goodfire's Ember
2026-05-22Goodfire announces SOC 2 Type II certification
2026-02-25Goodfire publishes research on harvesting activations from a trillion-parameter model
2026-02-05Goodfire raises $150M Series B led by B Capital at $1.25B valuation
2025-08Goodfire partners with Radical AI for materials science interpretability work
2025-04-17Goodfire raises $50M Series A led by Menlo Ventures, including Anthropic's first startup investment
2024-12Goodfire launches Ember, its flagship interpretability API
2024-08Goodfire raises $7M seed round led by Lightspeed Venture Partners

Revision history

29 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

American Airlines Group Wayfair Safe Superintelligence Inc.