THE DAILY BRIEFING

Bittensor Daily — November 5, 2026

With #General and #Rao both falling back to seven messages apiece after Nov 3's busier moderation day, the week's governance and tokenomics threads did not resurface, leaving the protocol release and the subnet cards to carry the day's substance. SN118 (Ditto) pushed out another bench version that stalled its review pipeline, SN58 (Attune) miners accused others of copying crowned models, and the X feed carried both a spotlight on Trishool's safety model and a governance-structure claim about Subtensor itself. The day lacked any new tokenomics debate of the kind that ran through #Rao earlier in the week.

Subtensor merges a 33% combined cap on subnet burn/recycle emission slots alongside a new 'Gamma' dollar-valued service credit minted by burning Alpha.
(1) The emission-slot change lets subnet owners set nonzero burn and recycle shares as long as the two stay under a combined 33% cap, fixing a bug where sparse weight updates could zero out unspecified slots, and requires a spec_version bump from 482 to 483. (2) Gamma gives subnet teams a new tool to price and sell services in dollar terms by burning Alpha, with mechanisms for depositing, spending and publishing against the credit. (3) Both changes are marked breaking, meaning node operators and developers integrating with either mechanism must update, while TAO holders not interacting with either feature are unaffected. No community reaction to either change appears in today's channels.
Opentensor and Yuma Group spotlight Trishool's HaloGuard 2.0 as a state-of-the-art AI-safety result built on Bittensor.
Bittensor subnet @trishoolai just released a SOTA model for AI response safety. Their 0.8B HaloGuard 2.0 outscored Alibaba’s 4B and 8B Qwen3Guard-Stream models on average across six response-safety benchmarks. Trishool's HaloGuard Output Guard for AI safety and alignment is now #1 on the benchmark, outperforming much larger models from @nvidia, @Google, @alibaba_cloud, and @Meta. (1) Both of Bittensor's highest-profile accounts credit a single subnet's result the same day, a notable concentration of attention. (2) The claim centers on a smaller model (0.8B) beating larger named competitors (4B and 8B), though neither post offers independent verification of the benchmark methodology. (3) No dissenting or skeptical reply to either post appears in today's material.
SN118 (Ditto) ships Benchmark v16, raises its submission fee, and sees its review pipeline stall for hours.
(1) The 'dueling bench' update added miner-built generator cases, cost-based scoring, and a secrecy/audit scheme, activating once the top-5 cohort completed v16 quorums, but the pipeline then stalled, leaving most agents rejected or stuck under review. (2) The team raised the submission fee from 0.05 TAO to 0.10 TAO effective immediately and flagged a forthcoming breaking shift to a cost-first incentive model. (3) Two agents were rejected for an 'I7 breach' after a secondary-model veto was found blocking tool calls before dispatch; the token fell 5.4% over 24 hours, continuing a run of near-weekly tokenomics and scoring changes on the subnet.
SN58 (Attune) miners accuse others of copying crowned models by adding noise before resubmitting, and the team admits no safeguard exists yet.
(1) Miners pointed to specific repository timestamps as evidence of a workflow that downloads a top model, adds minor noise, and resubmits it to claim the 'king' crown. (2) The team acknowledged the concern and announced upcoming changes: submissions will stay private until a model is crowned king, and periodic Hugging Face availability checks will be removed. (3) No automated anti-copy system exists yet on the subnet, and no fix had shipped by day's end; the token rose 10.38% over 24 hours despite the dispute.
A #Rao member's dereg-price question gets two answers that differ in emphasis, while #General and the rest of #Rao stay near-silent.
How does current dereg price work? `tao_in_pool / alpha_staked` or `tao_in_pool / alpha_staked + alpha_in_pool`? drew The latter for subnets registered after June 4th 2026, before a second reply added both have protocol alpha in the denominiator. (1) The two answers are not stated as contradicting each other, but one ties the formula to a registration-date cutoff while the other stresses a shared denominator term, and neither is explicitly reconciled. (2) This is a narrow mechanism question rather than a policy debate, unlike the week's emission-decay and SN82 threads, and the asker was satisfied after the first reply. (3) #General's 7 messages and #Rao's 7 messages both continued the sharp falloff from Nov 3's 27 and 23, with routine moderation, greetings, and a single SN124 (Swarm) promotion filling the rest of the day.

#General

1 REPORT

Other

Other: a job-solicitation reminder, routine greetings, the standing scam warning, a PFP rule reminder, and a shared Zils post filled a near-silent day

Opening · 05:41–23:08 UTC

A moderator opened the day repeating the channel's standing rule that No solicitations for jobs or partnerships are allowed here, before the rest of the day's seven messages ran to greetings -- Hello everyone drew Hello Gideon -- and a sign-off wishing Happy Thursday everyone 🫡.

Moderation · 15:00–15:18 UTC

The channel's full anti-impersonation reminder ran again, including the warning that People claiming to be mods, admins, Opentensor Foundation members, subnet support, or any kind of "support" sending you a direct message or friend request -- they're imposters, block them! and that The "Subtensor Wallet" and "Dsperse Wallet" are both SCAMS, alongside a separate profile-image reminder that Solid-color PFPs that mimic standard Discord color themes (or which are transparent) are not allowed in this server.

Link share · 18:20 UTC

A member posted an outside link to a post introducing Zils, Subnet 127 on Bittensor, describing it as training custom System One models, without further discussion in the channel.

3 things to read from it

Volume fell sharply to 7 messages today from Nov 3's 27, returning the channel to the pattern of very low, non-substantive traffic seen on Oct 31, Nov 1 and parts of Nov 2 rather than continuing Nov 3's busier moderation day.
No ecosystem-level question, mechanism dispute or governance topic reached #General today; the only subnet mention was a passive link share with no reply.
The Zils (SN127) link echoes the identity-change question #General fielded about the same subnet on Nov 3, but today's post drew no discussion of that earlier friction.

In the week’s context

On Nov 3, #General logged questions about SN127 (Zils)'s on-chain identity change that were redirected to archived channels rather than answered directly; today's link to the same subnet drew no comparable discussion.
On Nov 2, the channel's standing scam warning ran twice amid an otherwise quiet day; today's single repetition of the same warning continues that pattern of recurring, unprompted reminders rather than new incidents.

Where it stands

Closed as routine; nothing in today's #General traffic was left open for further discussion.

#Rao

2 REPORTS

Notable

A member asks how deregistration price is calculated, and two replies seem to differ before converging

Question · 13:48 UTC

A member asked plainly: How does current dereg price work? `tao_in_pool / alpha_staked` or `tao_in_pool / alpha_staked + alpha_in_pool`?

Reply · 14:17 UTC

One reply pointed to the second formula for a specific cohort: The latter for subnets registered after June 4th 2026

Clarification · 17:31 UTC

A second member then qualified that the distinction may not matter as much as it looks, noting both have protocol alpha in the denominiator

2 things to read from it

The two answers are not stated as contradicting each other but their emphasis differs -- one ties the formula to a registration-date cutoff, the other says both variants share the same denominator term -- and neither is reconciled further in the material.
This is a narrow mechanism question rather than a policy debate, unlike the week's emission-decay and SN82 threads, and it was resolved quickly with no pushback from the asker, who simply thanked the first reply.

Where it stands

Answered twice, with the asker satisfied after the first reply; the second, more technical clarification came after and was not explicitly reconciled with the first.

Other

Other: a member promotes SN124 (Swarm)'s drone-patrol demo, and the channel's only other traffic is a single unexplained mention

Promotion · 12:15 UTC

A member shared SN124's demo clip and urged others to pay attention, writing Show some love to swarm guys , actually doing things alongside a linked post describing a patrol run where 3/3 intruders spotted, 2 dogs ignored and 100% of the park searched.

2 things to read from it

This is the first #Rao mention of Swarm since Oct 30's report that its Sentinel challenge backlog was clearing, now framed as a positive showcase rather than a backend-strain complaint.
Total channel volume fell to 7 messages from 23 on Nov 3, continuing the sharp drop-off since Oct 31's 101-message peak, with no tokenomics or governance thread resurfacing today.

In the week’s context

Oct 29 and Oct 30 covered SN124 (Swarm)'s Sentinel challenge launch and its evaluation-queue backlog; today's post is the first #Rao mention since then, and it is purely promotional rather than a follow-up on the backlog.

Where it stands

Closed; a one-line show of support with no reply or further discussion.

Subnets

5 REPORTS

SN127 24h −18.05%

Zils

What happened: SN127 (Zils) introduced itself as a subnet training custom small "System One" decision models for specific customer use cases. The project runs nightly "soup" competitions in which miners commit model bodies, these are graded, and the best ones are combined and paid. The team confirmed the competition runs on mainnet, not testnet, and that outside validators are intended to grade rounds themselves. The official @zils_ai account posted the subnet's public launch announcement.
Community: Community members asked repeatedly about competition timing, grading deadlines, and whether committed bodies would be included in the next round, as well as when emissions to miners would begin. Some messages pointed out discrepancies between the website and the stated schedule, and the team acknowledged the site needed updating. Reaction to the official launch post was positive.
Official: The team said mining and emissions are live now, explained that each pool runs a nightly exam with bodies committed before 7 PM ET graded for the soup, and stated the website (zils.ai/soup and zils.ai/miners) was being updated to reflect current competition status.

SN70 24h −21.93%

Fugal

What happened: SN70 (Fugal) team answered miner questions on infrastructure, scoring, and TDX attestation requirements, confirming GCP as the only supported provider for now and detailing cost and accuracy scoring formulas, including how completion-token averages update from live exploration. The team outlined upcoming changes: a live leaderboard on fugalai.com showing miner accuracy, cost and payout, an updated model list, and additional TDX providers over time, with 24-hour notice to be given before changes and dates posted in the repo in advance.
Community: Miners asked repeatedly about using VPS options other than GCP, citing convenience, and the team said this is not a current priority. A community member proposed Phala Cloud as an additional TDX provider, citing compatibility with the existing dstack and PCCS setup and lower cost, which the team said it would look into; another community member later noted Phala Cloud's public IP limitation would require additional infrastructure work.
Official: The team said its current focus is subnet incentives and ensuring miners produce valuable routers, with plans to expand provider support after that, and confirmed the reveals are published publicly one epoch later via Google Cloud Storage.

SN118 24h −5.40%

Ditto

What happened: SN118 (Ditto) rolled out bench v16, the 'dueling bench,' adding miner-built generator cases, cost-based scoring factors, and a secrecy/audit scheme for generator cases; it activated once the top-5 cohort completed its v16 quorums. The subnet's review pipeline then stalled for several hours, leaving most agents rejected or stuck under review. The team raised the submission fee from 0.05 TAO to 0.10 TAO effective immediately and announced a forthcoming shift to a cost-first incentive model favoring low-cost, high-accuracy agents. Two submitted agents, crown-v208 and hfull-h3 v11, were rejected after a secondary-model veto was found to block tool calls before dispatch, described as an 'I7 breach.'
Community: Miners repeatedly asked when agents stuck in 'Under review' would be processed and why the leaderboard showed only a few agents. Some called the fee increase unfair to miners, while others speculated about operator availability during the stall.
Official: The team confirmed bench v16's activation at about 12:16Z, announced the upload fee increase to 0.10 TAO, and outlined a breaking direction change toward a cost-first incentive model. It also published rejection reasons for crown-v208 and hfull-h3 v11, citing a confirmed I7 breach in each case.

In debate · SN118 review pipeline stall and operator availability · 11 participants

Miners reported the review process stuck for hours, with agents left pending and no response from operators.

“it seems like system stuck on something please check and fix as soon as possible, thanks”
Some community members joked or speculated that the reviewing operator was unavailable, attributing the outage to the operator itself.

“Peyton's assistant was in a commuter plane crash 😢 RIP”
Others pointed out that only a handful of agents remained on the leaderboard and suggested miners were not participating due to low emissions.

“it appears they are not participating, likely because subnet emissions are low”

SN105 24h −2.31%

Beam

What happened: Miners in the SN105 (Beam) channel reported that room-transfer tasks were repeatedly going to a small set of orchestrators (around 12-15) at short, recurring intervals, with one miner citing data showing 2,714 of 2,847 chunks from a batch of transfers going to just 15 orchestrators. After an investigation, BeamCore confirmed the flagged room transfers are excluded as self-dealing and said the anti self-dealing system in PRISM is being upgraded, with new room-transfer test scenarios planned. Separately, the team declined to restore full historical dashboard and telemetry data publicly, citing the risk that competitors could train on it. Beam also reported moving 1TB of data in 2 minutes 42 seconds, 20% faster than last month, while running 2.5x more security checks.
Community: Miners repeatedly questioned why the same small set of orchestrators kept receiving the room-transfer tasks and asked the team to explain or exclude the request; some also pushed for public dashboard data so miners could spot anomalies themselves. The team asked miners to stop tagging staff and to file bug reports instead of repeating accusations.
Official: BeamCore confirmed the flagged room transfers are excluded as self-dealing, said the anti self-dealing system is being upgraded, and announced new room-transfer test scenarios covering all eligible orchestrators as a time-limited optimization campaign.

In debate · SN105 (Beam) room-transfer task distribution dispute · 6 participants

Miners argued a fixed group of orchestrators in the Toronto room-transfer task was receiving tasks every few minutes in a way that looked engineered rather than organic.

“This 12 miners set by prisms core of the beamcore side. but it publish task every 3minutes to cheat”
The team pushed back on the repeated framing, asking for a formal bug report rather than continued accusations.

“4th time you are saying selfmining. Have you reported the bug? (if it is one)”
The team later confirmed the specific transfers were self-dealing and removed them from scoring.

“These transfers are now flagged as self dealing and excluded from PRISM”

SN58 24h +10.38%

Attune

What happened: Miners on SN58 (Attune) accused other participants of copying top-performing models by adding minor noise before resubmitting, allowing the copies to claim the 'king' crown. One miner pointed to specific repository timestamps as evidence, and others described a workflow of downloading a model, adding noise, and resubmitting for evaluation. The team acknowledged the feedback and announced upcoming changes: miner submissions will remain private until a model is crowned king, and periodic Hugging Face availability checks will be removed.
Community: Several miners proposed solutions, including making only the king model public and verifying model similarity before accepting a new crown, with one calling private commits the best option. The dispute continued after a new king was crowned, with one miner on X publicly contesting the result as based on a copied model and calling the eval pipeline broken; others in the channel defended the subnet and asked critics to wait for the team's fixes.
Official: The team stated it is testing changes to improve the competition, including making miner submissions private until crowned king and removing periodic Hugging Face availability checks.

More subnet community coverage every day in the Telegram channel →

Deregistration price formula

A user asked how the current subnet deregistration price is calculated, specifically whether the denominator uses only alpha_staked or alpha_staked plus alpha_in_pool. Two other users responded with differing descriptions of the formula.

Positions

One proposed formula divides tao_in_pool by alpha_staked plus alpha_in_pool.

“tao_in_pool / alpha_staked + alpha_in_pool”
A community member said the alpha_staked plus alpha_in_pool version applies specifically to subnets registered after June 4th 2026.

“The latter for subnets registered after June 4th 2026”
Another community member said both versions of the formula actually include protocol alpha in the denominator.

“both have protocol alpha in the denominiator”

Official response: There has been no official response from the team or Opentensor in the messages provided.
What to watch: Whether Opentensor or subnet documentation issues a clear, authoritative statement on the exact deregistration price formula.

@MacrocosmosAI

Macrocosmos introduced IOTA Decide 4B, its first public model, described as trained across distributed GPUs and released with open weights under Apache-2.0.

View the original post on X ↗
@KoreanCryptoKOR

Targon, running on SN4 (Targon), reportedly generated $1.2M in revenue in October, its first month above $1M, with daily revenue said to exceed $40K.

View the original post on X ↗
@trishoolai

Trishool AI said its HaloGuard 2.0 model, built on Bittensor, outperforms Qwen3Guard-Stream across sizes as a 0.8B open streaming guard.

View the original post on X ↗
@TPN_Labs

TPN Labs released TPN-007 on Hugging Face, a 2.32 GB compressed version of Qwen3.5-4B for SN65 (Gradients), built to run on Gratis AI.

View the original post on X ↗
@chronollm

ChronoLLM, built on SN38, said its upcoming 16B MoE point-in-time model is nearing pre-training completion and claims improved scores over prior and comparable models.

View the original post on X ↗
@chutes_ai

Chutes AI announced DeepSeek V4.1 Flash is live on SN64 (Chutes), positioning it alongside V4 Pro for routing between lighter and heavier model tasks, citing mixed benchmark comparisons to Opus 5.

View the original post on X ↗
@taoperp

TAOPERP said it now allows long and short trading on TAO and SN64 (Chutes)'s token, with up to 10x leverage in USDC.

View the original post on X ↗
@byteleap_ai

Byteleap AI highlighted Grayscale's Oct 23 S-1/A filing describing Subtensor as Proof-of-Authority, with node participation requiring Opentensor Foundation approval and most nodes Foundation-controlled.

View the original post on X ↗

◆2026-11-05 · End of this edition

#3278 Breaking change Affects emission

Allow nonzero etao burn/recycle slots with 33% combined cap

What: Subtensor now permits subnets to set nonzero burn and recycle emission slots as long as the two combined stay under a 33% cap, and fixes a bug where sparse weight updates could zero out slots that were not explicitly specified.
Why: The change gives subnet owners more flexibility in how they allocate emissions between burning, recycling and distribution, while preventing an unintended side effect that could erase unrelated emission settings.
Who it affects: Subnet owners can now configure burn/recycle emission shares within the new 33% limit; ordinary TAO holders, validators and miners are affected indirectly through how a subnet's emissions are split, but take no direct action. The update requires a spec_version bump from 482 to 483.

#3300 Breaking change

Add gamma: dollar-valued service credits minted by burning subnet alpha

What: Subtensor adds a new 'Gamma' token: subnets can burn Alpha tokens to mint a dollar-valued service credit, along with mechanisms for depositing, spending and publishing services using it.
Why: The feature gives subnets a way to create a stable, dollar-denominated unit for paying for or pricing services, separate from Alpha's market-driven value.
Who it affects: Subnet teams gain a new tool to price and sell services in dollar terms by burning Alpha; miners and validators on subnets that adopt Gamma may see new payment or service mechanisms, while TAO holders not interacting with such subnets are not directly affected.

#3302

feat(ci): add targeted benchmark runs

What: A CI workflow update lets maintainers run targeted benchmark checks on selected pallets and functions instead of the full benchmark suite.
Why: This makes benchmark testing faster and more explicit for developers working on specific changes.
Who it affects: This is a tooling-only change with no runtime migration or spec-version bump; it affects only Subtensor developers and has no direct effect on TAO holders, validators or miners.

Past editions +

24H Market

Gainers

Losers

Bittensor Daily — November 5, 2026

#General

Other: a job-solicitation reminder, routine greetings, the standing scam warning, a PFP rule reminder, and a shared Zils post filled a near-silent day

#Rao

A member asks how deregistration price is calculated, and two replies seem to differ before converging

Other: a member promotes SN124 (Swarm)'s drone-patrol demo, and the channel's only other traffic is a single unexplained mention

Subnets

Zils

Fugal

Ditto

Beam

Attune

Deregistration price formula

Allow nonzero etao burn/recycle slots with 33% combined cap

Add gamma: dollar-valued service credits minted by burning subnet alpha

feat(ci): add targeted benchmark runs

24H Market

↗Gainers

↘Losers

Bittensor Daily — November 5, 2026

Today's highlights

Discord conversations worth following

Other: a job-solicitation reminder, routine greetings, the standing scam warning, a PFP rule reminder, and a shared Zils post filled a near-silent day

A member asks how deregistration price is calculated, and two replies seem to differ before converging

Other: a member promotes SN124 (Swarm)'s drone-patrol demo, and the channel's only other traffic is a single unexplained mention

Zils

Fugal

Ditto

Beam

Attune

Ecosystem debate

Deregistration price formula

Voices on X

Protocol

Allow nonzero etao burn/recycle slots with 33% combined cap

Add gamma: dollar-valued service credits minted by burning subnet alpha

feat(ci): add targeted benchmark runs

Recent editions

Gainers

Losers