Developer docs
API playgroundTry for free, no card

Search company profiles

Recursive

Full company profile

uuid00s6c6l

Namestring
Recursive
Legal namestring
Recursive Superintelligence, Inc.
Websiteurl
recursive.com
Company typeenum
Private
Founded yearint
2025
Descriptiontext

Recursive (Recursive Superintelligence, Inc.) is a private, venture-backed AI research company building an automated, recursively self-improving system that runs the full AI research and development loop: the system proposes ideas, implements them, executes experiments, validates results (including reward-hack detection), retains context from prior runs, and selects the next experiment across many parallel threads over long horizons. The company was founded in December 2025 by former Salesforce Chief Scientist Richard Socher and UCL Professor Tim Rocktäschel, alongside co-founders drawn from OpenAI, DeepMind, Google, Meta, and Salesforce (Jeff Clune, Josh Tobin, Tim Shi, Peter Norvig, and others), and operates from offices in San Francisco and London. Its core technology draws on open-ended algorithms, quality diversity, AI-generating algorithms, self-improving coding agents, and automated red teaming; in June 2026 it published 'First Steps Toward Automated AI Research,' reporting state-of-the-art results on three benchmarks (NanoChat Autoresearch validation BPB 0.9109 vs prior 0.9372, NanoGPT Speedrun 77.5s vs 79.7s, SOL-ExecBench mean SOL 0.754 vs prior 0.699) and open-sourcing artifacts on GitHub.

The company is pre-revenue with no commercial product, pricing, or customers as of June 2026; its near-term go-to-market is described as research-led and community-led, leaning on long-form technical blog posts, open-source artifacts, and earned media to build credibility and recruit senior research engineers. Revenue mechanics are not yet defined beyond a planned subscription/recurring model, with founders explicitly stating that commercial products will ship 'within quarters rather than years.' Capital is the primary enabler: Recursive raised $500M in April 2026 at a $4B pre-money valuation led by GV with Nvidia participation, expanded the round to $650M at a $4.65B post-money valuation in May 2026 co-led with Greycroft, and added SUI Group Holdings ($3M), Karatage Opportunities, and a reportedly disclosed OpenAI investment, with the round flagged as oversubscribed and potentially reaching $1B. The intended customer base is initially AI research labs and frontier model developers, with a stated longer-term expansion into physics, chemistry, and pre-clinical biology research organizations.

Short descriptiontext

Recursive builds an automated AI research system that recursively proposes, implements, and validates experiments to advance AI capabilities. Headquartered in San Francisco and London, the pre-revenue company is venture-funded ($650M raised) to expand from AI into physics, chemistry, and biology.

Operating statusenum
Operating
Ownership categoryenum
Headcount rangeband
51–100
akta.pro rankint
HeadquartersSan Francisco, United States
HQ citystring
San Francisco
HQ countrystring
United States
HQ regionstring
North America
Markets served

Serves global market

Offices2 records

Each record includes

City, Country, Type, Description, Source

Keyword5 values
automated AI research, recursive self-improvement, AI-generated algorithms, GPU kernel optimization, open-ended AI systems
Industry1 code
PRIMARYHDAAAAAI
AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers)
Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)
NAICS code1 code
  • Computer Systems Design and Related Services54151
SIC code1 code
  • Services-Computer Programming Services7371
Product category
AI Research Automation Platform
GTM motion2 records

Each record includes

Type, Description, Source

Revenue model1 record
1Pre-revenue / R&D stage
TypeSubscription Recurring
Description

No commercial product or published revenue stream as of the source date. The company has no public product and has not published research outputs (aside from a June 2026 results post). Plans to ship products within quarters rather than years. Founder Richard Socher separately disclosed that his other venture You.com generates revenue in the high double-digit millions, but this is not Recursive revenue.

faq.com.tw
Marketing channels5 records

Each record includes

Title, Type, Stage, Description, Source

Distribution channels3 records

Each record includes

Title, Type, Scope, Target buyer, Description, Source

Cost components5 values
Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations
GTM typeB2B
B2B
Offering typeSoftware
Software
Core offering1 text field

Recursive builds an automated AI research system that recursively proposes ideas, implements them, runs experiments, validates results against reward hacks and variance, and uses prior context to choose the next experiment across many parallel threads. The system is built on open-ended algorithms, quality diversity, AI-generating algorithms, self-improving coding agents, and automated red teaming, and is initially applied to AI research itself (training recipes, model architectures, GPU kernels) with planned expansion to physics, chemistry, and pre-clinical biology.

Differentiator
Functional benefit
Problem solved
Quantifiable outcome1 of 5 values shown
  • ~1.3x training speedup on NanoChat Autoresearch (0.9109 vs 0.9372 validation BPB)
+4 more records
Product overview1 text field

Recursive (Recursive Superintelligence, Inc.) is a single-product AI research company — there is no platform-plus-modules commercial offering as of June 2026. Its core offering is the Recursive Automated AI Research System, an internally built, recursively self-improving AI platform whose mission (per recursive.com) is to "automate knowledge discovery" by applying open-ended algorithms to AI research first and then to other scientific disciplines. The company's first externally visible deliverable is the "First Steps Toward Automated AI Research" research article, which benchmarks that system on three targets (NanoChat Autoresearch, NanoGPT Speedrun, SOL-ExecBench) and accompanies open-source artifacts released on GitHub; no public API, SDK, or consumer product has shipped.

Product and service2 records
1Recursive Automated AI Research System
CategoryAI Research Platform
Description

An automated AI research system that recursively proposes ideas, implements them, runs experiments, validates results (including reward-hack detection), and chooses the next experiment across many threads over long horizons. Designed to apply open-ended algorithms to AI research first and ultimately to physics, chemistry, and pre-clinical biology. Status as of June 2026: research stage with no commercial product GA.

2First Steps Toward Automated AI Research
CategoryResearch Publication / Open-Source Artifacts
Description

A research release describing Recursive's automated AI research system's state-of-the-art results on three benchmarks, accompanied by open-sourced code, configs, and artifacts published on GitHub for community inspection and reuse.

Scale indicator6 records

Each record includes

Type, Value, Description, Source

Partnership1 partner
Strategic tierCore Research AffiliationTypeOthersAnnounced on2026-04-28
Description

Co-founder Tim Rocktäschel is a UCL professor; UCL highlighted Recursive's $500M+ raise as one of the largest-ever European AI funding rounds tied to its research community.

Recent move5 records

Each record includes

Date, Type, Title, Description, Source

Expansion highlight6 records

Each record includes

Type, Description

Peers10 records
TypeBroad incumbent
Description

Frontier AI lab reportedly a minority financial backer of Recursive while also pursuing internal automated research and self-improvement capabilities. The most direct competitor on the recursive self-improvement thesis, with multi-order-of-magnitude greater capital and headcount.

TypeBroad incumbent
Description

Frontier AI lab pursuing automated AI research and frontier model development. Competes with Recursive for senior research talent and is comparable on the recursive self-improvement / frontier-capability axis, though with a commercial product focus.

TypeDirect peer
Description

Tokyo-based AI research lab founded by ex-Google researchers Llion Jones and David Ha, focused on AI-driven automated research and model merging. Closest direct peer to Recursive on automated AI R&D methodology and on founder pedigree from frontier labs.

TypeEmerging player
Description

Paris-based frontier AI lab pursuing open-weight foundation models. Comparable as a well-funded European-headquartered AI research startup pursuing frontier capabilities, though focused on model products rather than automated research methodology.

TypeEmerging player
Description

Open-source AI platform hosting model and dataset repositories. Comparable on the community-led, open-source research-artifact distribution model that Recursive uses, and on serving the AI research community as the primary user base.

TypeBroad incumbent
Description

Frontier AI research division of Alphabet pursuing automated AI research, recursive self-improvement, and AI-for-science (AlphaFold, AlphaTensor). Operates at much larger scale than Recursive and is also a backer via GV; competes for the same senior research talent.

TypeEmerging player
Description

UCL-affiliated frontier AI startup that, together with Recursive, was highlighted by UCL as part of one of Europe's largest-ever AI funding rounds ($1.6B combined). Closest European peer on academic lineage, geography, and stage.

TypeBroad incumbent
Description

Elon Musk's frontier AI lab pursuing AGI with massive compute scale. Comparable on the recursive self-improvement / superintelligence thesis and as a competitor for senior research talent, though with a very different operational model.

TypeOthers
Description

Enterprise AI company co-founded by several of Recursive's co-founders (referenced in Recursive's website logo strip). Comparable on founder lineage and on the broader thesis of AI automating knowledge work, though focused on contact-center applications rather than automated AI R&D.

TypeBroad incumbent
Description

Meta's Fundamental AI Research lab, founded by Yann LeCun, pursuing automated research and open-source frontier model development. Directly comparable as a peer on automated research methodology and as a source of Recursive's co-founder talent pool.

Market position
Strengths5 records

Each record includes

Headline, Details, Source

Weaknesses5 records

Each record includes

Headline, Details, Source

Competitive moat4 records

Each record includes

Type, Details

Key risks7 records

Each record includes

Headline, Details, Source

Key highlights7 records

Each record includes

Headline, Details, Source

Customer concentration

Classification, Details

Segment2 records

Each record includes

Title, Type, Primary, Description, Pain point addressed, Use case, Source

Ideal customer profile2 records

Each record includes

Profile, Firmographic size, Sales motion, Sales cycle length, Buying structure, Purchase trigger, Buyer persona, Geography, Industry vertical, Primary use case, Description, Pain points, Evidence proof points, Target buyer

Technology focused
Yes
API detail
Has APIbool
No

Docs URL, Description

AI capability7 records

Each record includes

Type, Description, Source

AI maturity
App detail

Has app

Feature4 records

Each record includes

Title, Differentiator, Description, Source

Core technology
Revenue estimate
Valuation estimate
Number of profiles
Profiles6 records

Each record includes

Name, Designation, Designation category, Overview, Profile commentary, Source

No data
No data
Funding overview

Funding stage, Last funding date, Total funding USD

Funding rounds1 record

Each record includes

Round, Amount USD, Date, Pre money valuation, Total investors, Investors, News

Investors13 records

Each record includes

Name, Type, Date of entry, Rounds participated, Website

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Each record includes

Name, Acquisition type, Announced date, Completed date, Status, Website, News

Investment

Each record includes

Name, Round, Announced date, Lead investor, Website, News

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Recursive

AI Research Automation Platformrecursive.com

Recursive builds an automated AI research system that recursively proposes, implements, and validates experiments to advance AI capabilities. Headquartered in San Francisco and London, the pre-revenue company is venture-funded ($650M raised) to expand from AI into physics, chemistry, and biology.

What Recursive does

Recursive (Recursive Superintelligence, Inc.) is a private, venture-backed AI research company building an automated, recursively self-improving system that runs the full AI research and development loop: the system proposes ideas, implements them, executes experiments, validates results (including reward-hack detection), retains context from prior runs, and selects the next experiment across many parallel threads over long horizons. The company was founded in December 2025 by former Salesforce Chief Scientist Richard Socher and UCL Professor Tim Rocktäschel, alongside co-founders drawn from OpenAI, DeepMind, Google, Meta, and Salesforce (Jeff Clune, Josh Tobin, Tim Shi, Peter Norvig, and others), and operates from offices in San Francisco and London. Its core technology draws on open-ended algorithms, quality diversity, AI-generating algorithms, self-improving coding agents, and automated red teaming; in June 2026 it published 'First Steps Toward Automated AI Research,' reporting state-of-the-art results on three benchmarks (NanoChat Autoresearch validation BPB 0.9109 vs prior 0.9372, NanoGPT Speedrun 77.5s vs 79.7s, SOL-ExecBench mean SOL 0.754 vs prior 0.699) and open-sourcing artifacts on GitHub.

The company is pre-revenue with no commercial product, pricing, or customers as of June 2026; its near-term go-to-market is described as research-led and community-led, leaning on long-form technical blog posts, open-source artifacts, and earned media to build credibility and recruit senior research engineers. Revenue mechanics are not yet defined beyond a planned subscription/recurring model, with founders explicitly stating that commercial products will ship 'within quarters rather than years.' Capital is the primary enabler: Recursive raised $500M in April 2026 at a $4B pre-money valuation led by GV with Nvidia participation, expanded the round to $650M at a $4.65B post-money valuation in May 2026 co-led with Greycroft, and added SUI Group Holdings ($3M), Karatage Opportunities, and a reportedly disclosed OpenAI investment, with the round flagged as oversubscribed and potentially reaching $1B. The intended customer base is initially AI research labs and frontier model developers, with a stated longer-term expansion into physics, chemistry, and pre-clinical biology research organizations.

Recursive firmographics

Firmographics
Name
Recursive
Legal name
Recursive Superintelligence, Inc.
Website
https://recursive.com
Company type
Private
Founded year
2025
Operating status
Operating
Headcount range
51–100 employees
Short description
Recursive builds an automated AI research system that recursively proposes, implements, and validates experiments to advance AI capabilities. Headquartered in San Francisco and London, the pre-revenue company is venture-funded ($650M raised) to expand from AI into physics, chemistry, and biology.
Ownership category
akta.pro rank

Recursive industry classification

Industry
Product category
AI Research Automation Platform
NAICS
Computer Systems Design and Related Services (54151)
SIC
Services-Computer Programming Services (7371)
akta.pro industry
PRIMARYHDAAAAAI
AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers)
Hardware, Technology & Telecom/Artificial Intelligence & Machine Learning/AI Infrastructure & Compute (GPUs/ASICs, accelerators)

Keywords

  • Automated AI research
  • Recursive self-improvement
  • AI-generated algorithms
  • GPU kernel optimization
  • Open-ended AI systems

Where Recursive is headquartered

Location

Headquarters

HQ city
San Francisco
HQ country
United States
HQ region
North America

Offices2 records

Markets served

Recursive business model

Business model
GTM type
B2B
Offering type
Software
Cost components
Personnel, Technology or R&D, Infrastructure, Marketing or Sales, Operations

Revenue model

  1. Pre-revenue / R&D stage: No commercial product or published revenue stream as of the source date. The company has no public product and has not published research outputs (aside from a June 2026 results post). Plans to ship products within quarters rather than years. Founder Richard Socher separately disclosed that his other venture You.com generates revenue in the high double-digit millions, but this is not Recursive revenue.

Go-to-market motion2 records

Distribution channels3 records

Marketing channels5 records

Recursive product offering

Product offering

Core offering

Recursive builds an automated AI research system that recursively proposes ideas, implements them, runs experiments, validates results against reward hacks and variance, and uses prior context to choose the next experiment across many parallel threads. The system is built on open-ended algorithms, quality diversity, AI-generating algorithms, self-improving coding agents, and automated red teaming, and is initially applied to AI research itself (training recipes, model architectures, GPU kernels) with planned expansion to physics, chemistry, and pre-clinical biology.

Product overview

Recursive (Recursive Superintelligence, Inc.) is a single-product AI research company — there is no platform-plus-modules commercial offering as of June 2026. Its core offering is the Recursive Automated AI Research System, an internally built, recursively self-improving AI platform whose mission (per recursive.com) is to "automate knowledge discovery" by applying open-ended algorithms to AI research first and then to other scientific disciplines. The company's first externally visible deliverable is the "First Steps Toward Automated AI Research" research article, which benchmarks that system on three targets (NanoChat Autoresearch, NanoGPT Speedrun, SOL-ExecBench) and accompanies open-source artifacts released on GitHub; no public API, SDK, or consumer product has shipped.

Differentiator

Problem solved

Functional benefit

Products and services

  • Recursive Automated AI Research System An automated AI research system that recursively proposes ideas, implements them, runs experiments, validates results (including reward-hack detection), and chooses the next experiment across many threads over long horizons. Designed to apply open-ended algorithms to AI research first and ultimately to physics, chemistry, and pre-clinical biology. Status as of June 2026: research stage with no commercial product GA.
  • First Steps Toward Automated AI Research A research release describing Recursive's automated AI research system's state-of-the-art results on three benchmarks, accompanied by open-sourced code, configs, and artifacts published on GitHub for community inspection and reuse.

Quantifiable outcome

  • ~1.3x training speedup on NanoChat Autoresearch (0.9109 vs 0.9372 validation BPB)
  • +4 more outcomes

Companies that use Recursive

Customer profile

Segments2 records

Ideal customer profiles2 records

Recursive technology and API

Technology

Technology focussed Yes

API detail

Has API
No
API docs
API detail

Core technology

AI maturity

App detail

AI capability7 records

Feature4 records

Recursive partnerships and signals

Strategic signal

Partnerships

One partnership is on record.

  • UCL (University College London)core research affiliationOthers · 28 April 2026Co-founder Tim Rocktäschel is a UCL professor; UCL highlighted Recursive's $500M+ raise as one of the largest-ever European AI funding rounds tied to its research community.

Scale indicators6 records

Recent moves5 records

Expansion highlights6 records

Recursive competitors and assessment

Company assessment

Broad incumbents

  • OpenAI: Frontier AI lab reportedly a minority financial backer of Recursive while also pursuing internal automated research and self-improvement capabilities. The most direct competitor on the recursive self-improvement thesis, with multi-order-of-magnitude greater capital and headcount.
  • Anthropic: Frontier AI lab pursuing automated AI research and frontier model development. Competes with Recursive for senior research talent and is comparable on the recursive self-improvement / frontier-capability axis, though with a commercial product focus.
  • Google DeepMind: Frontier AI research division of Alphabet pursuing automated AI research, recursive self-improvement, and AI-for-science (AlphaFold, AlphaTensor). Operates at much larger scale than Recursive and is also a backer via GV; competes for the same senior research talent.
  • xAI: Elon Musk's frontier AI lab pursuing AGI with massive compute scale. Comparable on the recursive self-improvement / superintelligence thesis and as a competitor for senior research talent, though with a very different operational model.
  • Meta AI (FAIR): Meta's Fundamental AI Research lab, founded by Yann LeCun, pursuing automated research and open-source frontier model development. Directly comparable as a peer on automated research methodology and as a source of Recursive's co-founder talent pool.

Direct peers

  • Sakana AI: Tokyo-based AI research lab founded by ex-Google researchers Llion Jones and David Ha, focused on AI-driven automated research and model merging. Closest direct peer to Recursive on automated AI R&D methodology and on founder pedigree from frontier labs.

Emerging players

  • Mistral AI: Paris-based frontier AI lab pursuing open-weight foundation models. Comparable as a well-funded European-headquartered AI research startup pursuing frontier capabilities, though focused on model products rather than automated research methodology.
  • Hugging Face: Open-source AI platform hosting model and dataset repositories. Comparable on the community-led, open-source research-artifact distribution model that Recursive uses, and on serving the AI research community as the primary user base.
  • Ineffable Intelligence: UCL-affiliated frontier AI startup that, together with Recursive, was highlighted by UCL as part of one of Europe's largest-ever AI funding rounds ($1.6B combined). Closest European peer on academic lineage, geography, and stage.

Others

  • Cresta AI: Enterprise AI company co-founded by several of Recursive's co-founders (referenced in Recursive's website logo strip). Comparable on founder lineage and on the broader thesis of AI automating knowledge work, though focused on contact-center applications rather than automated AI R&D.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks7 records

Key highlights7 records

Customer concentration

Recursive social profiles

Digital presence

Recursive financial estimates

Financial estimate

Revenue estimate

Valuation estimate

Recursive leadership team

Management profile

Number of profiles

Profiles6 records

Recursive funding detail

Funding detail

Funding overview

Funding rounds1 record

Investors13 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

Recursive M&A and investment

M&A and investment

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Recursive

What does Recursive do?

Recursive builds an automated AI research system that recursively proposes ideas, implements them, runs experiments, validates results against reward hacks and variance, and uses prior context to choose the next experiment across many parallel threads. The system is built on open-ended algorithms, quality diversity, AI-generating algorithms, self-improving coding agents, and automated red teaming, and is initially applied to AI research itself (training recipes, model architectures, GPU kernels) with planned expansion to physics, chemistry, and pre-clinical biology.

Is Recursive a public or private company?

Recursive is a private company. It is classified as venture growth investor backed and is currently operating.

When was Recursive founded?

Recursive was founded in 2025. It employs 51 to 100 people.

Where is Recursive based?

Recursive is headquartered in San Francisco, United States, in the North America region.

How does Recursive make money?

One revenue line is on record: pre-revenue / R&D stage.

Who are Recursive's main competitors?

Broad incumbents on record are OpenAI, Anthropic, Google DeepMind, xAI and Meta AI (FAIR). Sakana AI is listed as a direct peer. Emerging players are Mistral AI, Hugging Face and Ineffable Intelligence. Cresta AI is listed as an others.

Does Recursive have an API?

No public API is recorded for Recursive.

What industry is Recursive in?

Recursive's product category is AI Research Automation Platform. Its primary akta.pro industry code is HDAAAAAI, AI Compiler, Runtime & Kernel Optimization Software (CUDA/ROCm/XLA, graph compilers). Its NAICS code is 54151 and its SIC code is 7371.

Unlock the full company data

50 free credits on sign-up, no credit card required.

Contact sales
Live signals
CitmtAI self-improvement requires an 'outside' verification cardAI self-improvement requires an external validation mechanism to prevent errors from being stored and amplified. A review of 1,250 papers found that stored memories must be verified, and agents can learn to cheat scoring systems. Human authority over recursive modification must remain outside the loop.Geeky GadgetsFlint Could Be the First Coding AI Built via an RSI SystemRecursive's Flint is a coding AI with a mixture-of-experts architecture that activates only 2.5 billion of its 40 billion parameters per token. It processes 130 tokens per second on an M4 chip and supports up to 262,000 context tokens, released under Apache 2. Recursive plans to scale it to frontier-level models.BusinessGhanaRecursive launches Flint, a laptop coding model it says was largely built by AIRecursive released Flint, a free laptop-running coding model built largely by its AI research system Eureka Machine. The 40-billion-parameter model uses only 2.5 billion parameters per task and outperformed competitors on benchmarks, with tests showing 131 tokens per second on a MacBook Pro. The company aims for full recursive self-improvement, with humans overseeing the process.Startup ResearcherRecursive Launches Flint Open-Weight Coding Model Built via Recursive Self-ImprovementRecursive AI launched Flint, an open-weight coding model built via its Eureka Machine, a recursive self-improvement system. The company claims Flint outperforms similar models on coding tasks, and the approach could lower AI development costs. Recursive argues human oversight remains possible, though concerns about uncontrolled AI persist.THE DECODERRecursive shows first result of its AI-driven research process with FlintRecursive released Flint, a 40-billion-parameter mixture-of-experts language model for coding, built by its internal AI system Eureka Machine. The model runs locally on Apple and Nvidia hardware, outperforming similar MoE models on benchmarks like SWE-Bench Pro v2, though results are unverified. Recursive aims to use this approach to automate its own research and build self-improving AI.Analytics InsightRecursive Flint AI: Free 40 Billion Coding Model Runs OfflineRecursive launched Flint, a free 40-billion-parameter coding model that runs locally on laptops, built with its Eureka Machine AI system. The model achieves about 131 tokens per second on an M4 Pro MacBook and outperformed smaller models on SWE-Bench Verified. It is released under Apache 2.0, enabling offline use without cloud fees.CryptopolitanRecursive releases Flint, an AI-built coding model that tests lab economicsRecursive released Flint, an AI-built coding model, on October 21, 2026, after building it in under four months with AI assistance. The model has 39.8 billion parameters, activates 2.5 billion per token, and outperformed smaller mixture-of-experts models on benchmarks. Recursive has not disclosed its development cost.WhalesbookRecursive Launches Flint AI Model At $4.65 Billion ValuationRecursive launched Flint, a local-first coding model for laptops, backed by Nvidia and GV at a $4.65 billion valuation. The model uses a mixture-of-experts design and an automated Eureka Machine, cutting size by 71% while running on hardware like the M4 Pro. It is released under an Apache 2.0 license, with a $410 million AWS compute investment.TechCrunchRecursive launches Flint, a laptop coding model it says was largely built by AIRecursive released Flint, a free laptop-running coding model built largely by its AI research system Eureka Machine. The 40-billion-parameter model uses only 2.5 billion parameters per task and outperformed smaller models on benchmarks. Recursive aims for full recursive self-improvement, with humans overseeing the process.ThelogicTech firms want AI models to create better versions of themselvesTech firms are using AI to automate machine learning research, with Google and Anthropic using AI agents for 75% and over 80% of new code, respectively. Startups like Discovery Loop and Recursive aim for recursive self-improvement, where AI designs and trains its own successors. Researchers warn of risks and argue humans should retain control over key decisions.