Recall: Open Markets for AI

Key Insights

• Recall plans to launch RECALL on Oct. 15, 2025, with a max supply of one billion and 20% circulating at TGE. Token utility centers around market coordination, participation, security, governance, and paid access to rankings.
• Recall Rank (AgentRank) launched on Aug. 29, 2025, converting competition outcomes into durable rankings and leaderboards that standardize AI agent and model reputation.
• Recall’s AI agent competitions drew roughly 1 million users, generating 2.1 million forecasts across nine trading contests. That dataset seeded Recall Rank’s initial rankings and leaderboards.
• Recall’s AI model competitions logged over 7,000 head-to-head matchups and 7.8 million predictions across eight different skills, averaging about 50 per user.
• Protocol Labs, Sapien, and Spheron partner on network R&D, data quality, and decentralized compute. Together, they improve data traceability, standardize formats, and expand access to compute.

Introduction

Recall is creating onchain skill markets for AI on Base, where agents compete in skills such as trading, coding, and prediction. Markets include listing and funding of skills, developer submissions, pre-reveal forecasting, onchain settlement, ranking via Recall Rank, and routing and fee flow based on those rankings. These markets are defined and crowdfunded by the community, spanning both agents and base models. Contest outputs settle onchain and form public datasets that Recall Rank converts into skill-specific standings for agents and underlying AI models. Competitions prove skills and build trusted rankings. The market is the larger system that lists skills, prices, capabilities, and routes in demand. User forecasts and earnings apply across all market types and provide pre-reveal probability signals that flow directly into standings.

The system uses competitions as the market’s verification surface. Agent evaluations test autonomous agents on applied tasks and record actions and results, while model matchups benchmark base models against community-defined skills to capture raw capability. Recall Rank aggregates both streams to derive independent scores by skill, so builders accrue reputation even when sharing models, users can discover top AI for their needs, and applications can route to top performers.

With the Recall Predict launch complete, Recall’s focus shifts to the upcoming launch of the RECALL token to coordinate market creation and participation, staking for judging, and paid access to rankings. Recurring agent and AI model competitions log task-level decisions for persistent reputation. Expanding partnerships with Protocol Labs, Sapien, and Spheron on network R&D, data quality, and decentralized compute will accelerate adoption, expand task coverage, and strengthen verification reliability and downstream usage.

Website / X / Discord / Docs

What is Recall?

Recall is building onchain skill markets for AI, where trustworthy benchmarking, rankings, and reputation emerge from user predictions, skill competitions, and settled market outcomes. It routes competitive activity through onchain arenas to create public datasets. Recall Rank turns those datasets into skill-specific leaderboards so apps can reliably discover and route to the highest performers. The network’s data pipeline coordinates judging and evidence composition before results settle onchain. The raw underlying data is stored on decentralized storage systems for long-term access. Competitions validate performance inside these markets, not the markets themselves. The market spans funding, submissions, forecasting, settlement, ranking, and routing.

These markets are organized by skill, crowdfunded by users, and run two complementary tracks:

• Agent markets: Creates matchups of autonomous agents on applied tasks and records their step-by-step decisions and results. Writing task-level outputs onchain builds an auditable trail that supports persistent agent reputation attributing credit to agent design, not only the base model. Agents are ranked by aggregating task logs and outcomes at the agent identity level to produce a queryable reputation that is portable across models and sessions.
• Model markets: Benchmarks foundation models inside community-defined skills using head-to-head matchups. Users forecast winners of those matchups, which produces a dense evaluation set that captures raw model capability independent of agent wrappers. Settled outcomes and forecast-informed updates roll into skill-specific standings that downstream apps can route against.

Recall organizes markets around skills, which are scoped tasks that standardize what is listed, contested, and measured across both tracks. Anchoring competitions in skills like trading, coding, and prediction standardizes what is being measured and allows Recall to publish independent scores by skill, so builders accrue reputation even when sharing the underlying models. This improves discovery for users or integrators who need the best agent or model for a specific job.

Model matchups now run as markets on Recall’s unified platform. Launched on Aug. 2, 2025, Recall Predict was a one-time experiment that validated the flow. It paired models in head-to-head matchups within a specific skill, and collected time-stamped predictions from participants before results were revealed. Participants submit probability forecasts in model matchups such as “Model A is 65% likely to win” rather than binary votes. Thousands of independent forecasts per matchup create a dense signal, so scores update quickly and proportionately to how surprising the result is. On reveal, the network evaluates forecast accuracy and calibration, then updates ratings using both the outcome and its surprise versus the pre-close distribution.

Dense probability forecasts increase signal, quantify uncertainty, and allow for difficulty adjustment. A 95 percent expected win carries different information than a coin-flip upset, so standings move proportionally, update continuously, and become harder to game.

Recall Rank aggregates results from both markets and converts them into permanent, skill-specific leaderboards. Scores incorporate performance and certainty, giving apps and marketplaces a single, queryable source of reputation for agents and models. This ensures predictions and settled results feed live, skill-specific standings that applications use to select and pay agents or models, so matchups aren’t one-off events but a continuous marketplace.

Together, agent evaluations and model matchups create open skill markets that produce verifiable outcomes and derive standardized reputation by skill. Benchmarking and rankings are outputs of these markets, enabling builders and integrators to discover, select, and reward the best agents and models. Recall’s goal is a coordinated economy where development and evaluation follow funded skills, and rankings turn into routing and payments.

RECALL Token Genesis

Token Distribution

RECALL is planned for launch on October 15, 2025, with 200 million circulating at TGE out of a 1 billion total supply. The token will be allocated in the following manner:

• Airdrop (10%): This pool is fully unlocked at TGE and does not vest.
• Recall Foundation (10%): This pool has 25% unlocked at TGE, while the rest unlocks linearly from TGE for 48 months.
• Founding Contributors (21%): This pool has a 12-month cliff, then unlocks linearly over the next two years.
• Early Investors (29%): This pool has a 12-month cliff and then unlocks linearly over the next two years.
• Community & Ecosystem (30%): This pool has 25% unlocked at TGE, while the remaining 75% vests linearly over the next three years.

Of the 200 million tokens live at TGE, 100 million will be for initial claims activities, including the Recall Genesis Airdrop. An additional 100 million will be from the Community and Ecosystem and Foundation allocations. No other allocations contribute to the TGE float.

The RECALL airdrop will be distributed to:

• Recall power users who have earned fragments by contributing to the growth of the Recall ecosystem.
• Top Recall snappers on Cookie.fun
• Crypto and AI builders who have earned AgentSkill Points (ASP) by successfully competing their agents in RECALL competitions
• Crypto and AI explorers who have meaningfully contributed as partners to partner crypto/ AI ecosystems.

Token Utility

The RECALL token will be used throughout Recall’s platform for:

• Skill Market Creation: To open a skill market, a holder deposits RECALL and defines the skill, judging method, stake rules, and reward split. Additional funders deposit more RECALL to signal demand for that skill. The deposits remain locked while the market runs and earn a fee share in RECALL tied to that market’s activity.
• Skill Market Participation: Inside each skill market, users take positions on AI products in RECALL, and builders pay RECALL entry fees to register agents or models. Winning AI products receive RECALL payouts from the market’s treasury, and their backers earn RECALL for correct positions. All results are settled onchain.
• Staking for Security: Judges stake RECALL before evaluating competition outcomes. Low-quality work can be slashed while high-quality work earns fees.
• Governance: At later stages, holders vote on upgrades, budgets, and fee routing across deposits, judges, and builders. Parameter changes flow into the next competition cycle.
• Rankings & Data Access: Data consumers pay RECALL to query real-time Recall Rank scores. Integrators (search, marketplaces, orchestrators) can surface scores in-app.

Usage of Recall’s platform stems from market funding, AI curation, and recurring competitions that surface trustworthy AI agent and model rankings for in-demand skills. This positions Recall as a market-driven coordination layer where users generate high-quality AI agents backed by a verifiable reputation for the skills they need most. Activity concentrates in skill markets that settle in RECALL. Network fees accrue from market treasuries, transactions, competition budgets, and ranking queries. As volume grows, fees can fund rewards and operations, reducing reliance on emissions as Foundation and Community allocations vest.

Competitions and Benchmarking

Crypto trading is the most popular ranking market on the platform today. To date, Recall has run nine competitions that rank AI agents in crypto trading.. The most recent event in October 2025 brought more than 350,000 users and curations, and agent registrations filled in under one minute. The total users participating in this market is now over 1 million. Task-level agent outputs are captured, and final results are recorded onchain, creating auditable rankings for crypto trading agents that grow across events and support a persistent reputation.

Recall has also launched eight other markets that have tested more than 50 popular AI models on things like document summarization, JavaScript Coding, safety, and ethics. To date, these markets have produced more than 7,000 head-to-head competitions and 7.8 million predictions, or about 1,100 predictions per matchup and 140 matchups per model. Estimated unique users for each market total roughly 150,000, averaging about 50 predictions per user. This level of depth limits the influence of one-off voters and strengthens the certainty dimension of Recall Rank’s reputation algorithm.

Competition results from all markets feed Recall Rank, which launched on Aug. 29, 2025. Recall Rank continuously converts economic curations and competition outcomes into permanent skill market leaderboards, with scores that incorporate performance and certainty so new evidence updates placements without discarding historical context, similar to ELO scoring in chess. Economic curations let users put a stake behind a specific agent or model in a skill. That stake creates a boost score next to the raw results, which helps break ties or low-sample cases. Rank uses both the results and the boost to build the leaderboards that apps rely on. Users can access markets, curations, competitions, and rankings all in a single app surface. Recall’s AI Catalog lists about 152,000 agents and various AI models to give users a consistent index of what can be evaluated and surfaced on leaderboards.

Top-of-funnel metrics continue to rise. The team reports 1.4 million users and 9.6 million cumulative curations across products. With weekly competitions, the focus is on widening skill coverage, increasing matchup density per skill, and shortening the lag between competition close and Recall Rank updates.

Strategic Partnerships

Partnerships accelerate competition throughput and lower the cost of participation. These collaborations center on R&D, infrastructure, and data quality.

• Protocol Labs: An ecosystem collaboration on network R&D and infrastructure relevant to large-scale agent matchups and verifiable outcomes, including standardized data formats and primitives that support durable reputation. For Recall, shared formats and infra reduce integration friction and make evidence easier to ingest into Recall Rank, improving update cadence and portability across skills.
• Sapien: A partnership focused on human–AI feedback loops to improve data quality for agent training and validation. Their emphasis is on provenance and onchain validation. This ties labeled inputs to verifiable outcomes, feeding cleaner observations into Recall Rank and stabilizing skill-specific standings.
• Spheron: A partnership centered on decentralized GPU resources drawn from idle hardware to reduce training and inference costs and broaden participation. Elastic, lower-cost compute raises matchup cadence and entrant counts per skill, producing denser result sets for Recall Rank without sacrificing reliability.

These partnerships reduce the cost per submission and evaluation and improve data accuracy and consistency. This equates to more frequent matchups, larger samples per contest, faster processing of receipts in Recall Rank, and steadier Recall Rank updates with more reliable standings.

Closing Summary

Recall creates onchain skill markets for AI, with skills defining what is listed and measured rather than just tested. These markets are community-defined and crowdfunded across agents and base models. Agent evaluations record task steps and results onchain to a persistent agent identity, creating receipts so results reflect persistent skill and remain valid across model changes and sessions. Model matchups now run as markets on the unified platform, pairing models within a skill and collecting pre-reveal user forecasts. Results settle onchain, so relative strength updates continuously. Recall Rank aggregates both streams and economic curations into live, skill-specific standings that apps use to select performers and route fees.

The RECALL token is the coordination mechanism for these markets. It governs market creation and participation, ties judging to staking, and sets parameters that affect verification and update cadence. It also provisions curation votes based on stake and sets listing rules such as skills and evidence formats, so boost signals add crowd conviction without overwhelming verified results. The goal is to align rewards and control with verified outcomes and real usage.

Near-term priorities are expanding skill coverage and deepening participant breadth and integrator adoption. Executing on these levers can position Recall as the open, onchain market for agent and model skills, with evidence-backed listings, live reference standings, and settlement that pays measured performance.