THE DAILY BRIEFING

Bittensor Daily — October 24, 2026

Opentensor shipped Release 484, introducing lifecycle states for subnets and a new EVM precompile for liquid alpha consensus mode, alongside runtime update 483 covering EVM precompile input handling and proxy/utility call validation. Elsewhere, SN92 (REAX) saw a sharp price drop as its team rolled out eval v5, SN102 (ConnitoAI) faced community questions over a top-ranked miner's reproducibility, and SN34 (BitMind) paused submissions during a database migration.

01
Release 484 (spec_version 484) adds formal subnet lifecycle tracking and a new EVM precompile that lets smart contracts read and set a subnet's liquid alpha consensus mode.
This gives validators and wallet tools such as btcli more consistent stake and dissolution-state reporting, and lets subnet owners configure liquid alpha consensus directly from smart contracts.
02
SN92 (REAX)'s token fell 18.29% as the team confirmed eval v5 binds Oct 26 at 14:30 UTC and named uid 164 the new REAX Engine champion at 3.43x reference engine throughput.
The team attributed the losses to an intended equilibrium where inefficient miners exit while efficient ones remain profitable, rather than a protocol problem.
03
On SN102 (ConnitoAI), community members said they could not reproduce a top-ranked miner's reported scores and noted a reported SN92 alpha transfer from an owner wallet to that miner's coldkey, prompting the team to open an investigation.
The dispute touches on the fairness of validator scoring, which affects trust in the subnet's ranking system.
04
SN34 (BitMind) paused new submissions after a database refactor overloaded servers and delayed scoring of more than 40 submissions, while confirming it has fully moved to a new direction called Yugos.
The pause directly affects miners awaiting evaluation, and the Yugos shift marks a departure from BitMind's prior AI media detection focus.
05
Conjectures said its miners produced a Lean-verified proof of Erdős problem 354(i), a question that had been open since 1971.
It is presented as a concrete research output attributed to a Bittensor subnet's mining network.

SN92 24h −18.29%

REAX

What happened: The team confirmed that eval v5 binds from Monday, Oct 26, 14:30 UTC, after a second validation check, and published the eval v5 format, a 300-item dev set and the reference scorer in the reax-subnet repository. Validator release 1.2.0rc11.post4 went live, with other validators asked to upgrade. Generalist track places 2 and 3 emissions went live for uid 160 (17%) and uid 113 (8%) after a delay, and uid 164 was crowned new REAX Engine champion at 3.43x reference engine throughput, surpassing the previous champion uid 176 (3.22x). Decision Models (SN92) went live on New York Token Exchange via TrustedRouter.
Community: Miners discussed that roughly 80% of miners in the inference pool are losing money, with some miners reporting continued break-even or profitable results depending on their setup and GPU costs. Some miners asked about the status of the delayed place-2 and place-3 emissions before they went live.
Official: The team explained the losses as part of an intended equilibrium where inefficient inference setups exit while efficient ones remain profitable, and pointed miners toward efficient engines, owned or cheap hardware, and serving multiple models. The team also confirmed the eval v5 timeline, validator release, and emission start times for the generalist track.

SN58 24h +7.46%

Attune

What happened: A community member reported a bug in the SN58 (Attune) simulator where an object disappeared mid-duel after the gripper released it. The team reviewed the replay and attributed the issue to a rare glitch in the SAPIEN physics simulator, where excessive contact force on release can cause objects to disappear or pass through the table. The team said this does not affect scoring fairness since all models run under the same simulator conditions. Separately, Robotensor opened its Playground to the public, letting users watch live robot task simulations with customizable scenes and camera angles.
Community: Miners discussed whether the current top-scoring 'king' model had benefited from the simulator glitch, with one calling it lucky, and debated whether models should be re-evaluated given score deltas as small as 2.9. A miner also asked whether merging or adding noise to leading models, instead of training new ones, was the intended competitive approach on the subnet.
Official: The team confirmed the disappearing-object issue stems from a rare SAPIEN physics glitch rather than a scoring bug, said no action is needed from miners for now, and asked for duel links and scene IDs to investigate further cases. Robotensor's official X account announced the public launch of the Playground.

SN118 24h +2.21%

Ditto

What happened: The team identified a bug in DittoBench v13 in which case-file dates in some memory cases appear in shuffled order, causing the "trust the latest date" scoring rule to produce incorrect results; a generator fix is tracked for v14. Separately, miners reported multiple agent submissions stuck in "screening incomplete" status for an extended period. The team traced this to a screener issue, re-queued several stalled submissions, withdrew a rejection tied to a tool-result character cap, and raised miner compensation to 35%. Ditto also announced a credit-usage forecasting tool built with SN91 (Cascade).
Community: Miners repeatedly asked the team to review and retry stuck submissions, with some requesting the team's assistant be enabled when unavailable. One miner questioned the rationale for the current 2-day privacy window. Miners also proposed that off-validator evaluation runs include receipts such as code hash and dataset seed to allow reproducibility, and suggested token cost count toward scoring alongside time.
Official: The team confirmed the v13 date-shuffling bug and screener issue, re-queued affected submissions, withdrew a rejection reason, raised miner compensation to 35%, and said it is considering moving evaluation runs off validators and building a diff-based, family-aware screener.

SN102 24h −2.80%

ConnitoAI

What happened: Community members questioned the consistent top ranking of a miner operating under the name sharp0203 (publicly linked to an identity called putty), arguing that reproducing the model's results did not match the reported scores. A community member said running the public model with the same validation seeds used on SN102 (ConnitoAI) produced different results. Separately, a message referenced a reported SN92 (alpha) transfer to the same miner's coldkey from an owner wallet, which some read as a sign of special treatment. The team, through its community manager, opened an investigation into the reproducibility claim and the validator scoring.
Community: Multiple members pressed for the miner's identity and an explanation, some calling the pattern of consistent first-place finishes suspicious. Other members pushed back, defending the miner's standing and the alpha transfer as unrelated to unfair treatment, while one self-identified alpha holder asked the channel to stay technical rather than emotional.
Official: The community manager asked the member who reproduced different results to share cycles, seeds, commands and outputs so the team could attempt to reproduce the discrepancy, and pointed to public code and a public validator dashboard. The team relayed the miner's own explanation that his lead comes from incremental gains on an 8×H200 setup, and later clarified that two disputed validator scores came from different rounds and were not directly comparable.

In debate · Alleged unfair miner dominance on SN102 (ConnitoAI) · 13 participants

A miner said reproducing the top performer's model with identical validation seeds did not reproduce the reported results, raising doubts about the scoring.

“He keeps coming in 1st every single time, but right now that just doesn’t add up.”
The team responded that the code and validator evaluations are public and verifiable by anyone, rather than relying on the team's word.

“The code is public. It's on github.com/Connito-AI. The results are public. The dashboard shows each validator's evaluation for every miner, so anyone can compare.”

SN34 24h +0.12%

BitMind

What happened: A database refactor on SN34 (BitMind) overloaded servers and left validators stuck, delaying scoring of more than 40 submissions. The team paused new submissions to complete pending runs and finish a database migration before reopening submissions. Separately, SN34 has fully moved to Yugos, a new router-like direction similar to JEV, replacing its prior AI media detection focus.
Community: Miners reported long stalls in validator scoring and asked why scores decreased and costs increased after re-evaluation, questioning the fairness of the process. Other community members urged patience and said the team was working hard to resolve the issues.
Official: The team said a major database refactor intended to improve the competition put heavy load on the database, causing runs to stall or fail, and outlined a plan to pause new submissions, finish current runs, migrate the database, then reopen submissions. The team also confirmed the Yugos launch, saying SN34 has fully moved to Yugos.

In debate · SN34 validator stalls and scoring fairness · 7 participants

A miner questioned why validators stalled specifically when a submission threatened to dethrone the current champion and why scores and costs shifted after re-evaluation.

“Why do validators get stuck whenever a submission has the potential to dethrone the current champion?”
The team acknowledged the frustration and attributed it to runs not finishing or being graded in a timely manner, pledging to get all runs evaluated.

“the root cause of frusteration is from the runs not finishing and being graded in a timely manner”
Another community member defended the team's effort and urged others not to push them, citing their track record.

“They are trying very hard and I am really surprised they made this result.”

More subnet community coverage every day in the Telegram channel →

@conjectures_io

Conjectures reported that its miners produced a Lean-verified proof of Erdős problem 354(i), a question that had been open since 1971.

View the original post on X ↗
@reliquary_ai

Reliquary AI said it has started a reinforcement learning run to train Teutonic I, a 9B-parameter model, with progress tracked on a public dashboard.

View the original post on X ↗
@CryptoZPunisher

A community member highlighted SN51 (LIUM), citing over $30 million in hosted GPU hardware, 225 NVIDIA B300 GPUs, and 91% capacity rented out according to the team.

View the original post on X ↗
@byteleap_ai

A community member noted the first on-chain eTAO emission splits: SN8 (Vanta) set owner cut to 0%, SN41 (Almanac) moved from 18% to 10%, and SN36 and SN70 took the 33% maximum.

View the original post on X ↗
@opentensor

Opentensor said eTAO lets subnet owners fund compute and research more freely, with a subnet owner on Novelty Search outlining plans for subnets to trade services and for builders to borrow against alpha.

View the original post on X ↗
@robotensor_ai

Robotensor announced its Robotensor Playground, where users can watch a robot arrange a room, is now open to everyone.

View the original post on X ↗
@IOTA_SN9

SN9 (IOTA) said it ran 7 independent 16B pretraining runs of about 140 GPUs each, with 585 worker crashes, preemptions and rejoins, and reported nearly identical loss curves to demonstrate reproducibility.

View the original post on X ↗
@CryptoZPunisher

A community member said that, per a chart, TAO allocated to subnets grew sharply from February 2025 through spring 2026 but has since plateaued, now standing at 17.63%.

View the original post on X ↗

◆2026-10-24 · End of this edition

New release · v484 Released

Subtensor v484: continuous stake reporting during Subnet dissolution, larger Root claim limit, and EVM Liquid Alpha controls

This release keeps Stake visible and consistently reported while a Subnet is being dissolved, and fixes Root claims for Validators whose Basket holds all 128 Subnets plus TAO. Subnet Owners can now read and set the Liquid Alpha consensus mode through the EVM Subnet precompile. Metagraph displays in the CLI now show the correct token symbols.

Read the upgrade report →

#3279 Breaking change

Release 484

What: This release (spec_version 484) adds formal lifecycle tracking for subnets (registered, started, pending dissolution, dissolving) with consistent stake and position reporting across that lifecycle, and adds a new EVM precompile function allowing smart contracts to read and set a subnet's liquid alpha consensus mode.
Why: The previous reporting could show gaps or double-counting of stake positions around payout and dissolution events, and liquid alpha consensus mode was not previously accessible to EVM-based tooling.
Who it affects: TAO holders and validators get more accurate, consistent stake and dissolution-state reporting in wallets and tools such as btcli and the Python SDK; subnet owners building EVM integrations can now read and configure liquid alpha consensus mode directly from smart contracts; miners are not directly affected.

#3280 Breaking change

release-483

What: Runtime 483 updates how EVM precompile inputs are handled and how proxy and utility call transactions are validated.
Why: The change improves correctness and robustness of transaction validation for these call types, supported by added documentation and test coverage.
Who it affects: This mainly affects developers building on EVM precompiles or using proxy/utility calls; ordinary TAO holders, validators, and miners are not directly affected.

#3281

perf(swap): skip unchanged balancer storage writes

What: This is a performance optimization to the protocol liquidity adjustment logic so that storage is only written when the underlying balancer or reservoir values actually change.
Why: The previous implementation performed unnecessary database writes even when calls did not change any stored state, adding avoidable overhead.
Who it affects: This is an internal efficiency improvement with no change to liquidity behavior, accounting, or fees; it affects node performance only and is not noticeable to TAO holders, validators, or miners.

Past editions +

24H Market

Gainers

Losers

Bittensor Daily — October 24, 2026

REAX

Attune

Ditto

ConnitoAI

BitMind

Subtensor v484: continuous stake reporting during Subnet dissolution, larger Root claim limit, and EVM Liquid Alpha controls

Release 484

release-483

perf(swap): skip unchanged balancer storage writes

24H Market

↗Gainers

↘Losers

Bittensor Daily — October 24, 2026

Today's highlights

Subnet Community Pulse

REAX

Attune

Ditto

ConnitoAI

BitMind

Voices on X

Protocol

Subtensor v484: continuous stake reporting during Subnet dissolution, larger Root claim limit, and EVM Liquid Alpha controls

Release 484

release-483

perf(swap): skip unchanged balancer storage writes

Recent editions

Gainers

Losers