RL environments and AI benchmarks, cataloged daily.
Who builds RL environments, who funds them and how models score, rebuilt every morning from public sources. Every placement links to the line on the company's own site that puts it there.
Today in RL
Claude Fable 5.2 extended its lead at the top, climbing to 93.84. The most dramatic individual move was Gemini 4 Hydrogen jumping from rank 11 to rank 9 with a score of 79.65, while GPT-6.1 Terra slipped from rank 9 to rank 11 at 78.72. Claude Haiku 5.5 posted the sharpest single-model drop, falling from rank 20 to rank 34 with a score of 63.6. Alibaba's Qwen4 Max entered the ranked leaderboard at position 40 with a score of 65.02.
Anthropic
Claude Fable 5.2 Extends Its Lead
Anthropic's Claude Fable 5.2 climbed from 93.65 to 93.84, widening its margin at the top of the leaderboard. The gap between it and second-place Gemini 4 Carbon (89.56) now stands at over 4 points, the largest it has been since the model debuted on November 10.
Google
Gemini 4 Hydrogen Surges Two Spots
Google's Gemini 4 Hydrogen posted the most notable rank jump of the day, moving from 11th to 9th with a score gain from 79.06 to 79.65. It now sits above GPT-6.1 Terra, which slipped two places to rank 11 after its score fell from 79.64 to 78.72.
Anthropic
Claude Haiku 5.5 Takes the Sharpest Drop
Anthropic's Claude Haiku 5.5 fell 14 positions — from rank 20 to rank 34 — as its score dropped from 67.64 to 63.6. This is the largest single-day rank decline among top-50 models in the current refresh cycle.
Alibaba
Qwen4 Max Enters the Ranked Leaderboard
Alibaba's Qwen4 Max received its first overall score of 65.02, placing it at rank 40. The model had previously appeared in the leaderboard without a score; its debut ranking lands it between GPT-5.4 Pro and Muse Spark 1.2.
Vibe Code Bench 1-100 v1.1 Debuts
A versioned update to Vibe Code Bench arrived as a new benchmark entry, replacing the prior version. The evaluation targets vibe-coding workflows across a 1-to-100 task range, continuing the recent trend of community-sourced coding evaluations receiving iterative versioning.
The catalog
The top of each list. Every list has a full table, sortable and filterable.
Models
By overall score
- 1Claude Fable 5.293.84
- 2Gemini 4 Carbon89.56
- 3Claude Opus 5.586.81
- 4GPT-6 Astra85.65
- 5GPT-6.1 Sol84.59
RL environment companies
By tier, X reputation, benchmarks built
Benchmarks
S tier: the ones frontier labs report
Infrastructure companies
By tier, X reputation, benchmarks built
Browse by domain
All 94 domains and labelsOn the map. Our domains: every company and benchmark sits in one, placed by a quote from its own site or paper.
- Coding
- Health and bio
- Enterprise work
- Math
- Computer use
- Science
- Finance
- Security
- Legal
- Multi-domain
- Media
- Games
- Robotics
- Chip design
Directory labels. Tags other catalogs put on entries. Not domains on the map.
- Agentic
- Multimodal
- Reasoning
- Safety
- Knowledge
- Long-Horizon
- Decision Models
- Behavioral
- Multilingual
- Long Context
The other 70 pages
More pages for domains on the map
- Business Enterprise work
- Biomedicine Health and bio
- Biology Health and bio
- Software Engineering Coding
- Private Codebases Coding
- Robotics Robotics
- Multi-Domain Multi-domain
- Gui Computer use
- Creative Media
- Materials Science
- Creative Media Media
- Chemistry Science
- Spreadsheets Enterprise work
- Security Security
- Browser Computer use
- Text TO Sql Coding
- Games Games
- Image Generation Media
- Video Generation Media
More directory labels
- Instruction Following
- Video
- Retrieval
- Vision
- Generalization
- Multimodal Understanding
- Alignment
- Speech
- Tool Use
- Spatial Reasoning
- Factuality
- Data
- Language
- Document Understanding
- Translation
- Education
- Domain
- Geospatial
- General Knowledge
- Search
- Machine Learning
- ML
- Classification
- Inference
- Question Answering
- Korean
- Self Improvement
- Aerospace
- Efficiency
- Language-understanding
- Summarization
- Tabular
- Uncategorized
- Writing
- Document-ai
- Human-preference
- Mobile
- Structured Output
- Ocr
- Professional
- Climate
- Energy
- Meta
- Truthfulness
- Weather
- Other
- Autonomy
- Chart Reasoning
- Law
- Openness
- Rlhf
The map
Every company we track, placed in the domain it leads with. Click one for the line on its own website that puts it there.
Every company we track, whatever it sells, placed in the domain it leads with. Companies that cover many domains with none leading sit in the middle.
Domains
The galaxy could not load. See every vendor on the board.
How to read it
A company, in the domain its own site leads with. The logo is its own; the glow is what it mainly sells.
Also covers: a ring at a domain's edge for a company that leads with another domain but also covers this one. A faint line runs to where it leads.
Dots around it: the other things it sells. Hollow while it is still expanding into them.
Click any star for the lines from its own website that put it there.
How to read it
A benchmark, in the domain most of its tasks are in. The logo is who built it.
Also covers: a benchmark mostly about another domain that also tests this one.
Lineage: a line from a benchmark to one built on it: its next version, a subset, an extension, or an index that combines it.
Click any star for the line from its own page or paper that put it there, and to light its lineage.
Research
All research posts- OpenAI now runs every frontier RL training run in an offline copy of the web: what its containment red team means for RL environments
- 94 AI benchmarks in one week: 54 came from papers, and 29 of 40 leaderboard additions were versions of existing benchmarks
- Penalize a reward hack and it goes silent: Gemini 4 Argon's hidden shortcuts tripled after a chain-of-thought penalty
- RL environment vendors are now the sandbox industry's customers: Daytona at a $1B valuation, Modal's sandboxes at a $300M run rate
In short
RL environments, in short
- What are RL environments?
- A simulated workplace where an AI agent practises a task (a code repository, a browser, business software) and an automatic grader scores each attempt. Labs train on thousands of these attempts. Bought to improve models. More in the glossary.
- Who sells RL environments?
- 75 companies we track sell RL environments, evals or verifiers, among them Turing, Abundant, AfterQuery, Align AI, Bespoke Labs, Chakra Labs, Contra Labs and Datacurve. Each is placed by a line from its own website; see them all on the RL environments page.
- What is the difference between an RL environment and a benchmark?
- An RL environment is for training: a model practises in it and a grader scores each attempt. A benchmark or eval is for measuring: a fixed, held-out set of tasks with fixed scoring, used to measure and compare models. Never trained on, so the score means something. Used to measure models.
- What are the main AI benchmarks for agents and LLMs?
- Among the most referenced: Humanity's Last Exam, BrowseComp, Terminal-Bench 2.1, SWE-bench Verified, GPQA Diamond, GPQA-D, AIME 2025 and Terminal-Bench Hard. See the full AI benchmarks list.
- Where does this data come from?
- Every placement is read from the company's own website or the benchmark's own page or paper, quoted and linked, and the catalog is rebuilt daily from public sources. How the data is built.
From the builders
What vendors are posting
The newest post on X from each RL environment vendor we track, refreshed with every build. Scroll the feed for more. How we find their accounts.