Ridges is an Bittensor subnet that acts as an open source agent competition platform, where miners both compete and collaborate on a software engineering agent.
SN 62. Rank 31 of 32 by emission
Ridges
Emission1.0%of the block reward · rank 31 of 32
Participants13 miners · 11 validators
Registration0.02598 TAO
Verified2026-09-09links checked and scores merged
The programagents · github.com@6ada4b7 · read 2026-09-19
The work
A software engineering agent: one Python file exporting agent_main that takes a problem statement and returns a unified diff. The miner pays a burn of about $5 per upload and its own inference. What is paid for is an approved improvement over the competition's current leader on a hidden problem set within a niche such as linting or database queries.
own docs · docs.ridges.ai
agents submit a patch, the platform applies the diff and runs the hidden test suite
How it is scored
Deterministic: apply the patch, run the hidden tests, score the fraction that pass. Two screeners of 20 problems at 45 and 60 percent, then three validators on 50 problems each; a problem counts only when every assigned validator marks it solved. Test names, logs and inference results are hidden from miners. Each competition carries its own thresholds and problem counts.
own docs · docs.ridges.ai
the platform computes a consensus score: for each problem, the agent gets credit only if every assigned validator marks that problem solved
How the pay splits
Proportional to decaying reward scores across every approved agent, summed per hotkey. An agent qualifies only by beating the leader by 3 percent on score, or 6 percent on cost at equal score. Its score is improvement units, log-compounded 3 percent steps, times a time multiplier 1 + sqrt(hours stalled / scale), then halves every 336 hours. Each competition holds a fixed share.
code · raw.githubusercontent.com · utils/incentives.py
return value * 2 ** (-max(0.0, elapsed_hours) / half_life_hours)
Sources disagree The incentive page says the multiplier doubles after a 6-hour stall; api/config.py at the pinned commit defaults INCENTIVE_TIME_MULTIPLIER_SCALE_HOURS to 12, and each competition's policy may override it.
Who judges
Eleven validators were active on the 2026-09-09 chain snapshot. Commit-reveal is off. Validators run the agent in sandboxes on their assigned problems, but the weight vector itself comes from the platform: each validator fetches a hotkey-to-weight mapping from /scoring/weights, checks it sums to one and every hotkey is registered, and submits it unchanged. If any hotkey is missing, on-chain weights are left as they were.
chain · hyperparams-2026-09-19.json
commit_reveal_weights_enabled False, weights_version 6; chain-2026-09-09.json active_validators 11
How often it turns
Emission settles every 360 blocks, about 72 minutes. Validators fetch and set weights every 600 seconds by the example configuration. The puzzle turns per approval: a leader stands until an agent clears the bar, one submission per hotkey every 12 hours, and every reward halves over 14 days. A new registration is immune from deregistration for 7,400 blocks, about 25 hours.
chain · hyperparams-2026-09-19.json
tempo 360, immunity_period 7400
Who buys
Developers with a GitHub repository, through Ridgeline: one credit turns one issue into one pull request from an isolated machine. $9.99 a month buys 10 credits, or locking 1,000 alpha on the subnet earns one credit every three days. No usage or customer count is published; the third parties in the map cite cost per benchmark run, not paying customers.
own docs · docs.ridges.ai
A recurring subscription of $9.99/month grants 10 credits.
What has been gamed
Hardcoding: agents were written to pass known tests, so test names, logs and inference results are now hidden and problem sets are built and mutated per niche. Copying the leader and adjusting earns nothing below the 3 percent threshold. The rules allow optimising for a niche and ban stored answers. Agents whose hotkey deregisters are dropped before the split.
own docs · docs.ridges.ai
Miners previously hardcoded agents to pass known specific tests.
3 of 7 fields rest on the chain or the code. A manifest is the site's reading of the subnet's program, not a review; each field links its source. All programs · What a field means · Correct a field
What it says
The subnet in its own words: the line the index took from its front door, and what the team has said on the record. All 13 quotes in the map.
Their subnet
An open source competition where miners submit Python agents, validators run them in Docker sandboxes on benchmark problems, and emissions flow to approved agents by how much they beat the leader; the best agents are sold as Ridgeline.
a decentralized, self-sustaining marketplace of autonomous software engineering agents, designed to solve real-world software challenges.
we want to basically split each step up and get the miners to specialize in one kind of step... and then we basically create like a handoff layer where, when someone gives us a problem, we can split it up into all these steps and allow the agents to just focus on their one step
Ridges is an Bittensor subnet that acts as an open source agent competition platform, where miners both compete and collaborate on a software engineering agent.
Validators pull submitted code and run it on benchmark problems, evaluating the output. The highest-scoring agent earns emissions.
Ridges is evolving AI agents to be better and better able to solve software engineering problems end-to-end.
Bittensor, in their words
Bittensor appears only as the blockchain whose currency pays miners; the docs and README do not describe the network beyond that, and the product pages do not name it.
Your reward (emissions in the currency of Ridges Subnet on the Bittensor blockchain) is based on how much your agent improves...
Frames competition marketplace evolution competition platform handoff human developer leader niche path software engineer
What is said about it
By other voices in the map, verbatim and dated.
Most Bittensory subnet there is. Code is intelligence man! Unlocks open source models, wide open, winner-take-all, @chutes_ai & @TargonCompute integration, UID-less mining. What else do you need? $TAO
Subnet 62 Ridges is on a tear the last month and has now cracked the top 15 subnets. Join me and owner Shakeel on this pod as we jam on all things 62. This is one of my favorite new projects on Bittensor and has a great chance to disrupt AI coding.
Subnet 62 (Ridges) achieved 80% on SWE-Bench in under 45 days with minimal capital, showing Bittensor incentives unlocking efficiencies orders of magnitude beyond centralized incumbents.
Running all 500 SWE-bench problems with the top Ridges agent costs just $1.26, compared to $94 for Claude Opus.
It's five to seven times cheaper than Claude code and Codex at a comparable benchmark
What it shows
Explained from zero by the index, from the subnet's own materials, for someone who has never heard of Bittensor.
SN62 Ridges runs an open tournament for robot software mechanics: builders enter programs that fix broken code, the best fixer earns the prize, and the mechanics are hired out to developers at $9.99 a month for ten repairs.
The commodity, explained from zero
The mechanic is a software engineering agent: a program that reads a description of a bug or a feature request, edits the code, and hands back a patch. Ridges sells it through Ridgeline: connect a GitHub repository, open an issue, and the agent returns a pull request from an isolated machine. "1 credit -> 1 GitHub Issue -> 1 Pull Request." A subscription of $9.99 a month grants 10 credits; alternatively, locking 1,000 of the subnet's tokens earns one credit every three days.
The customer is a developer with a GitHub repository. The roadmap says the subnet launched in late 2024 as AgenTao, that the Ridgeline beta arrived in early 2026, and that in mid 2025 its agents "reached state-of-the-art performance among open-source SWE-bench agents", SWE-bench being a standard test of fixing real GitHub issues. No score accompanies that claim. No usage or customer figure for Ridgeline appears anywhere.
Why it is on Bittensor at all
The docs' argument is against closed agents: "Closed agents may also take the shortcut of hardcoding known solutions. Ridges removes that most basic of cheating strategies structurally." Problems are hidden and randomized, several independent validators must agree, and once an agent stops earning, "the exact agent.py that earned the score is open-sourced". The comparison breaks here: a mechanics' tournament has a sponsor who pays the purse, while the purse here is emission, the subnet's share of newly minted TAO, and the mechanic's tools (the language model calls) are paid for by the builder, at $10 to $20 per evaluation run.
How the work gets done
Miners (the builders) submit one Python file exporting agent_main, which takes a problem statement and returns a patch, paying a burn of about $5 per submission with one submission per hotkey every 12 hours. Validators (the judges) apply the patch and run a hidden test suite: two screening rounds of 20 problems at 45 and 60 percent pass thresholds, then three validators on 50 problems each, with credit only when every validator agrees. An agent earns emission by scoring at least 3 percent above the leader (a leader at 0.50 needs 0.515) or by being 6 percent cheaper at the same score, with a multiplier that doubles after a 6 hour stall and a 14 day half-life on every agent's reward.
How you would know it works
The leaderboard at ridges.ai/agents shows two open competitions dated September 9 and August 18, 2026, but only in a browser; the site returns a Vercel security checkpoint to anything else. No dated benchmark table with numbers exists in the materials.
What is missing
The chain contact hello@ridges.ai appears on no docs page fetched, and the site front page cannot be read by a fetcher. There is no API reference; Ridgeline is a GitHub app and a task board. Support is a FAQ and Discord, sales are final, and there is no status page or SLA. The docs and the chain identity link different Discord invites. Demand for Ridgeline is stated nowhere.
Go deeper
Sources for this explainer
- https://docs.ridges.ai
- https://docs.ridges.ai/incentive-mechanism.md
- https://docs.ridges.ai/ridgeline/credits.md
- https://docs.ridges.ai/ridges/ridges-way.md
- https://docs.ridges.ai/guides/mining-intro
- https://docs.ridges.ai/scoring.md
- https://docs.ridges.ai/roadmap.md
- https://raw.githubusercontent.com/ridgesai/ridges/main/README.md
Metaphor: an open tournament for robot software mechanics. Every claim is drawn from the evidence set or the subnet's own materials; "(inferred)" marks a conclusion rather than a quote. Corrections.
How legible it is
Four audiences, four questions each, rated on what a first-time reader can find in five minutes. The scale measures findability, not quality.
On-chain identity 6/7- subnet_name
- Ridges
- github_repo
- github.com
- subnet_contact
- hello@ridges.ai
- subnet_url
- ridges.ai
- discord
- discord.gg
- description
- Software Engineering Agents
- additional
- not set
Stakers and validators●●●●● 3.5
Should I allocate here? · rank 9 of 32 for this audience
Q1What it is●●●●● 4
You can use our state of the art agents to solve your GitHub issues today with our product, Ridgeline.
Agents and pull requests, one credit each, docs root, site front page animates its text
Q2Who it is for●●●●● 2
Check out the current leaderboard and status of active competitions at https://www.ridges.ai/agents
two open competitions dated Sep 9 and Aug 18 2026 in a browser, no Ridgeline usage or customers
Q3How it resists gaming or fails●●●●● 4
named page with 3 and 6 percent thresholds, a time multiplier table and a 14 day decay, anti gaming on the Ridges Way and scoring pages (hidden problems, screening judge), no failure modes
Q4Identity and documentation●●●●● 4
identity 6 of 7, github resolves, url renders in a browser but 429 to fetchers, credits page tells stakers how locked alpha earns credits, validator guide, contact not in the docs
Miners●●●●● 4.5
Can I compete, and what wins? · rank 1 of 32 for this audience
Q1What it is●●●●● 5
Your agent.py must export a single function: def agent_main(input: dict) -> str
taking a problem_statement and returning a unified diff, with an environment variable table and example code
Q2Who it is for●●●●● 4
Upload fee. A flat amount of Alpha burned from your registered wallet at the time of submission. Currently ~$5 per submission.
Inference $10 to $20 per run, beat the leader by 3 percent
Q3How it resists gaming or fails●●●●● 4
test pass fraction, screener thresholds 45 and 60 percent, worked numbers on the incentive page (0.50 needs 0.515, 2x at 6 hours), 12 hour cooldown, deregistered hotkeys dropped, no immunity policy
Q4Identity and documentation●●●●● 5
setup, local testing and submit pages match the ridges CLI in the README, release v0.3.2 published 2026-09-09, repo pushed the same day
Buyers and enterprises●●●●● 3.3
Can I use this today? · rank 14 of 32 for this audience
Q1What it is●●●●● 4
Provide a complete specification of the desired result, and the agent computes the code solution from a securely isolated compute environment
then a pull request, no example or endpoint
Q2Who it is for●●●●● 5
A recurring subscription of $9.99/month grants 10 credits.
One credit per pull request or 1 credit per 1,000 alpha locked per 3 days, customer is a developer with a GitHub repository
Q3How it resists gaming or fails●●●●● 2
FAQ, a privacy and security page and Discord, sales are final, no status page, no SLA, no response time
Q4Identity and documentation●●●●● 2
no API reference (Ridgeline is a GitHub App and a task board), chain contact hello@ridges.ai appears on no docs page fetched, Discord is the named channel
Newcomers●●●●● 3.8
What is this and why does it matter? · rank 9 of 32 for this audience
Q1What it is●●●●● 4
Ridges is evolving AI agents to be better and better able to solve software engineering problems end-to-end.
Plain, on the docs root rather than the site front page, matches chain
Q2Who it is for●●●●● 4
Closed agents may also take the shortcut of hardcoding known solutions. Ridges removes that most basic of cheating strategies structurally
hidden problems and open sourced code, no cost comparison
Q3How it resists gaming or fails●●●●● 3
roadmap claims state of the art among open source SWE-bench agents (Q3 2025) with no numbers, competitions page shows live leaderboards only in a browser, no dated benchmark table
Q4Identity and documentation●●●●● 4
subnet_name Ridges matches the site, description Software Engineering Agents is a tag, url resolves in a browser (429 to fetchers), the Ridges Way and roadmap pages serve as about
Provenance
How this rating came to be. Verified means the scorer fetched the page and the words are on it; secondhand means concluded from code, absence, or a third party. The link badge is a separate automated check made before publication.
scored by agent c 2026-09-09, rubric v1.0merged 2026-09-09links verified 2026-09-09: 37 live, 1 unreachable, 1 manual
Something wrong? Corrections of fact are applied as they arrive during the window; rating disputes are batched at its close. How to file one · GitHub issue · email.