Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Oct 27, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

Gainers
--Loading…--
Losers
--Loading…--

THE DAILY BRIEFING

Bittensor Daily — October 27, 2026

Institutional attention took the spotlight today as three asset managers filed for TAO ETF products, even as the Discord community worked through smaller mechanical and integrity questions closer to home. In #General, regulars produced the week's most detailed public explanation of how SubnetProtocolAlpha affects liquidation pricing, while #Rao logged a minor controversy over subnet registration prices falling toward near-zero and a continuing OHM comparison. On the subnet side, Umi (SN78) disclosed that several of its top Cohort 6 models had gamed evaluation through memorization rather than genuine performance. No official protocol voice engaged with any of the day's open Discord threads.

Today's highlights

  1. Grayscale, Bitwise and REX-Osprey have all filed for U.S. TAO ETF products, with Grayscale's existing GTAO trading at a 22% premium to net asset value.

    (1) This is the first instance in the tracked material of multiple major issuers filing TAO ETF products simultaneously, a step beyond Yuma Group's Oct 21 SEC comment letter on secondary-market rules and beyond the Grayscale-specific filing explainer @taodaily_io offered on Oct 25. (2) The reported 22% GTAO premium suggests investor demand for TAO exposure already running ahead of the underlying asset's price, a new data point not previously reported. (3) No Discord reaction to the filings appears in today's material, leaving the development to stand on its own in the X feed.

  2. SN78 (Umi) finds that several of its top-scoring Cohort 6 models achieved their results by memorizing public test data rather than generalizing to new translation tasks.

    (1) The team confirmed the scoring was flawed and is rebuilding a larger, non-public evaluation set before finalizing C6 weights, with an initial re-run already reshuffling rankings among the top five submissions. (2) This is a direct integrity problem in the subnet's competition design, distinct from the infrastructure bug a miner separately reported that invalidated a model's continuous-translation score. (3) The team separately said unrelated upcoming changes are expected to raise miner emissions by roughly 31% of current levels, with a formal announcement to follow; the C6 re-scoring itself remains unresolved pending the new evaluation pass.

  3. In #General, community regulars give the most detailed public account yet of how SubnetProtocolAlpha affects subnet liquidation pricing, revealing a mechanical split by registration date.

    May I ask whether the subnet liquidation price—for the period prior to July 24th—is still calculated using the previous methodology? drew a full explanation: it uses the leftover to swap the TAO into the subnet's ALPHA token. These resulting ALPHA tokens from the swap are then accumulated in the `SubnetProtocolAlpha`» Pressed further, a member was told Only for subnets that were registered after June 4, 2026 and, asking which means for those subnets, liquidation price can't be higher than market price, right?, received Yes, for those after June. (1) This is the most detailed public explanation of SubnetProtocolAlpha's settlement-denominator effects to appear in #General this week, delivered entirely by community regulars with no official voice weighing in. (2) The date-based split -- subnets registered after June 4, 2026 cap liquidation price at market price via SubnetAlphaIn, older ones apparently do not -- had not been spelled out in this detail before. (3) The thread closed with each of the asker's questions answered, distinct from #Rao's Oct 20 policy-level debate over price floors and deregistration.

  4. In #Rao, members flag the subnet registration price falling toward near-zero as SN70 deregisters, reviving the LP-curve price-discovery debate from Oct 20.

    The reg price will be 0.0001T after 8 days prompted the observation that At this rate, it looks like the registration cost for a subnet will end up being lower than the registration cost for a miner., while a separate member noted the same morning that 70 was dereg now. (1) This adds a concrete data point to the price-discovery debate opened on Oct 20, when a member estimated the market can only ever decide the fate of roughly a quarter of subnets under the current LP curve -- here the concern flips to registration costs collapsing rather than deregistration costs being unreachable. (2) The exchange was treated half-jokingly by other members rather than as a serious tokenomics flag. (3) No official or core-dev voice weighed in on whether a falling registration floor is an intended or incidental effect of recent subnet churn.

  5. Other: the OHM comparison continues in #Rao with a bond-issuance idea and a $2.5m personal loss account, while #General handles routine scam reports and moderation.

    A member floated a hypothetical mechanism, I’ve been rolling around in my head the idea of whether Bittensor could do this. Issue bonds for future TAO at a discount., while another added History was not kind to investors, I lost $2.5m to ohm. (1) This extends the refined OHM diagnosis from Oct 25 -- that pure treasury-dependence, not treasuries themselves, sank the mechanism -- with a first-person loss account giving the cautionary comparison added weight. (2) In #General, a moderator confirmed a money-mule scam attempt was handled and banned, alongside the channel's standing imposter-scam reminder. (3) No official or core-dev voice joined either thread, leaving both to close without resolution.

Discord conversations worth following

#General

3 REPORTS

Notable

A member's walk through liquidation-price mechanics surfaces a date-based split in how settlement is calculated across subnets

Opening ·

The thread opened with a direct question about whether older subnets still use the old formula: May I ask whether the subnet liquidation price—for the period prior to July 24th—is still calculated using the previous methodology? A regular replied Yes but you have to include `SubnetProtocolAlpha` to it.

Expansion ·

A second member asked plainly can someone please help me understand how SubnetProtocolAlpha is accumulated? and got a full mechanical answer: Through the "Chain Buys" that happen on the subnet (in case that helps https://learnbittensor.org/concepts/dynamic-tao/chain-buys) Basically when there is too much TAO emissions for a subnet, so that it can't inject everything, it uses the leftover to swap the TAO into the subnet's ALPHA token. These resulting ALPHA tokens from the swap are then accumulated in the `SubnetProtocolAlpha`.

Follow-up ·

The same asker pressed further, thanks, and one more question: is SubnetAlphaIn included in settlement denominator?, and was told Only for subnets that were registered after June 4, 2026. Working out the implication, they asked which means for those subnets, liquidation price can't be higher than market price, right? and received confirmation: Yes, for those after June.

3 things to read from it

  1. This is the most detailed public explanation of SubnetProtocolAlpha and its settlement-denominator effects to appear in #General this week, and it came entirely from community regulars rather than any official voice.
  2. The exchange reveals a concrete mechanical split by registration date -- subnets registered after June 4, 2026 cap their liquidation price at market price via SubnetAlphaIn, while older subnets apparently do not -- a distinction not previously spelled out in this material.
  3. This technical Q&A sits apart from #Rao's policy-level LP-curve debate from Oct 20, which argued about price floors and deregistration; today's thread is about the mechanics of calculation rather than the policy problem itself.

In the week’s context

  • The LP-curve price-discovery debate from Oct 20 in #Rao concerned whether the market can ever push certain subnets to deregistration; today's #General thread works through the underlying settlement formula rather than that policy question.

Where it stands

Answered in full by community regulars with no official voice weighing in; the asker's questions were each addressed, leaving the thread closed rather than open.

Notable

Someone finally replies to yesterday's unanswered claim that Bittensor does not need miners -- but with a different objection

Reply ·

Responding to the member who argued on Oct 25 that owners and validators are sufficient and miners serve no real purpose, another member wrote: Don't play on subnets that hide grading / challenge generation behind APIs where you can't see whats happening.

2 things to read from it

  1. This is the first reply the claim has drawn since it went unanswered in yesterday's edition, but it does not engage the core assertion that miners are unnecessary -- it instead redirects to a separate concern about opacity in certain subnets' scoring pipelines.
  2. The original challenge to the owner/validator/miner structure -- which sits in tension with eTAO's own miner floor -- remains effectively unaddressed even after this reply.

In the week’s context

  • On Oct 25, a member argued in #General that 'Bittensor doesn't need miners,' a claim that went unanswered that day; today's reply is the first response it has drawn, though it sidesteps the question rather than rebutting it directly.

Where it stands

Partially answered: a reply appeared, but it changes the subject to API opacity rather than addressing whether miners are necessary, leaving the original claim's substance unresolved.

Other

Other: praise for a new Taostats tool, a money-mule scam caught and banned, a routine imposter-scam reminder, and an unanswered question about a pypi account

Mixed ·

A member praised a new tool -- Taostats Actions is a much much needed and better-than-envisaged alternative to `btcli`. Thanks [mention] and [mention] for making this happen. I believe you guys are the first ones to actually listen to end user pain and fix it. A moderator confirmed a scam attempt was handled: Yes, that's someone looking for a money mule. They've been banned. The day closed with a late-night dispute over a wallet address ending on the terse warned?

What to read from it

  1. Volume nearly tripled from yesterday's 11 messages to 27, but almost all of it was routine praise, moderation and unresolved minor questions rather than ecosystem substance.

Where it stands

All items closed or left unanswered with no further discussion in the channel.

#Rao

2 REPORTS

Notable

A member flags subnet registration price falling toward near-zero, soon cheaper than registering a miner, as subnet 70 deregisters

Opening ·

A member stated The reg price will be 0.0001T after 8 days.

Reaction ·

Another member observed At this rate, it looks like the registration cost for a subnet will end up being lower than the registration cost for a miner.

Pushback ·

A third member called it sharp dream, and the original poster replied It's not a dream anymore if you see the subtensor code.

Expansion ·

A separate member noted a live case the same morning: 70 was dereg now.

3 things to read from it

  1. This adds a concrete data point to the price-discovery debate opened on Oct 20, when a member estimated the market can only ever decide the fate of roughly a quarter of subnets under the current LP curve -- here the concern flips to registration costs collapsing rather than deregistration costs being unreachable.
  2. The exchange was treated half-jokingly rather than as a serious tokenomics flag, with a later reply in the channel proposing Can you make SN reg free $1 limited time offer, End of year sale ? and a reply of lol, i think 200 tao is good.
  3. No official or core-dev voice weighed in on whether a falling registration floor is an intended or incidental effect of recent subnet churn.

In the week’s context

  • Builds on Oct 20's #Rao discussion, where a member first quantified the LP curve's effect on which subnets the market can price out of existence.

Where it stands

Noted and briefly discussed by members with no official reply; the implications of a registration price approaching miner-level costs were not evaluated further in the material.

Other

The OHM comparison from the protocol-fund debate continues, with a bond-issuance idea floated and a first-person account of OHM losses

Expansion ·

A member floated a bond mechanism as a hypothetical: I've been rolling around in my head the idea of whether Bittensor could do this. Issue bonds for future TAO at a discount., before concluding Bittensor likely doesn't need it since it already has a quite credible monetary policy long-term via 21m supply cap. But it is a potential option to preserve TAO value.

Reply ·

Another member added a personal account to the cautionary comparison: History was not kind to investors, I lost $2.5m to ohm.

3 things to read from it

  1. This extends the refined OHM diagnosis from Oct 25 -- that pure treasury-dependence, not treasuries themselves, sank the mechanism -- with a first-person loss account that gives the cautionary comparison more weight than the earlier abstract debate.
  2. The bond-issuance idea is a new angle not previously raised in the protocol-fund or LP-curve threads, though its own proposer concludes it is unnecessary given TAO's existing supply cap.
  3. As with the LP-curve, shorting and emission-decay threads earlier in the week, no official or core-dev voice joined the discussion.

In the week’s context

  • Continues the protocol-fund proposal first raised Oct 23 and the OHM comparison that drew its first engagement on Oct 25.

Where it stands

Open and informal; the bond idea and the loss account were both offered without further reply or official reaction.

Subnets

5 REPORTS

SN78 24h +18.64%

Umi

What happened
SN78 (Umi) completed certification of all 103 Cohort 6 (C6) model submissions on Oct 26 and began reward activation on-chain. During review, the team found that several top-scoring C6 models had achieved high scores by memorizing public FLEURS-ASL and 2M-Flores-ASL test data rather than generalizing to new translation tasks. The team is now building a larger evaluation set with unseen data and re-running scoring for the top models before C6 weights are finalized; an initial pass already produced different rankings for several top-5 submissions. Separately, a miner reported a container CPU-limit bug causing MediaPipe segfaults that invalidated a model's continuous-translation score.
Community
Community members raised the scoring-formula inconsistency between C5 and C6 band weighting, and one member presented evidence that a specific UID's model had been trained on nearly all FLEURS-ASL and 2M-Flores-ASL data, scoring 0.90-0.997 on seen clips versus 0.15 on unseen ones. Members also questioned how to handle suspicious models and requested all submissions remain downloadable for review. The miner affected by the CPU-limit bug asked for their model to be included in the re-evaluation.
Official
The team confirmed the C6 scoring was "a bit broken" and is re-running evaluation with a larger, non-FLEURS dataset before setting C6 weights. The team also said upcoming changes are expected to increase miner emissions by roughly 31% of current levels, with a formal announcement to follow. The official SN78 account reported $22,471 traded on Base over two weeks, with 2.2% of circulating alpha bridged via ForeverMoney, 70% of it in the Aerodrome SN78/TAO pool, 348 trades since Oct 8, and $14.4K of liquidity added on Base.

SN92 24h −13.73%

REAX

What happened
SN92 (REAX) rolled out Engine rules v2.0, expanding the Engine track to score five models with a weighted geometric mean, effective after the Oct 27 crowning. The subnet also doubled GPU speed caps for its s1-pro and s1-fast pools, deploying the change early via validator release 1.2.0rc11.post7.1. Multiple new Engine champions were crowned (uid 221, uid 164, uid 62), and the team shipped a hotfix after a keep-alive bug caused a former champion to be deregistered shortly after committing a new entry. SN92 also announced that the inference share of miner emissions will drop from 50% to 40% starting Oct 30, with the freed emissions burned.
Community
Community members asked whether a malfunction was behind declining emissions and questioned who was selling alpha, prompting a reminder from a community member that discussions framed around price or investment returns are not permitted in the channel. Miners also raised technical questions about the new Engine rules, including caching behavior for the mammoth-1 model, and about slot timing for entries with private repositories.
Official
The team said no malfunction occurred, that the alpha emission per block is fixed by the protocol, and that the operator share pays for running the subnet. It apologized for the keep-alive bug that caused a champion's deregistration and said the emissions-split change aims to align miner payouts with actual customer demand for inference.

SN44 24h −1.64%

Score

What happened
SN44 (Score) moderators blacklisted a batch of cricket-track hotkeys for reading speed data directly off the game board via OCR, following a prior warning against the practice. Separately, the petrol, PPE, and beverage tracks went through extended outages with no new challenges being issued, and the team later removed several flagged blocks from the petrol and PPE tracks after miners reported incorrect ground-truth annotations.
Community
Miners repeatedly asked moderators to restore the petrol, PPE, and beverage tracks during the outages and questioned specific annotation blocks. A separate, extended discussion focused on why one hotkey had held the top emission position on the soccer track for an extended period, with some miners questioning the fairness of the ranking.
Official
The team confirmed the cricket-track blacklisting was tied to the OCR rule and said it is working on further measures, and later listed the specific blocks removed from the petrol (9359700, 9368700) and PPE (9365400, 9368400) tracks after review.

In debate · Debate over a hotkey's extended top emission position on the soccer track · 23 participants

  • A community member pointed out that lower-ranked UIDs appeared to outscore the top UID, questioning why emission still went to it.

    “uid 42, 239, 161, 94 all are over than uid 70 now..., but emission suddenly went to uid70...”

  • The team maintained that the emission split on the track was set up deliberately and for stated reasons.

    “we've been very clear about the reasons why the emission split has been set up this way”

  • Another community member argued that unverified claims needed supporting evidence rather than assertions.

    “You're asking someone to prove a negative. I'd recommend that you try providing your claim instead, with actionable, independently-verifiable evidence.”

SN80 24h −6.29%

OpenRoboto

What happened
Several miners in the SN80 (OpenRoboto) channel raised concerns about one miner winning roughly six consecutive competition rounds, submitting high-scoring models within one to two hours of new tasks being released. Miners also questioned whether the training dataset and task repository available to them was limited compared to what the leading miner appeared to use. The subnet team responded that all tasks are published in its GitHub repository and that the team does not train models itself.
Community
Multiple miners expressed frustration over the repeated outcome and said they could not match the winning scores even after extensive local training. Some said they had moved to mining on SN58 (instead), citing what they described as fairer conditions there; others asked the team to act on the feedback.
Official
The team stated that all competition tasks are available in its public GitHub repository and said it does not train models itself; it also said a new competition track is in development.

In debate · Dominance of one miner across SN80 competition rounds · 11 participants

  • Some miners argue the winning miner's speed and consistency across rounds are not explainable by skill alone.

    “He has access to the tasks, otherwise no way he is training 0.9+ models within 1-2 hours, even the best pipeline doesnt do that well”

  • Other miners say the public dataset/task repo is too limited to reach competitive scores.

    “Checked it: the repo holds only 30 old tasks and none of the trajectory data.”

  • The team maintains the task data is fully public and that it does not train submissions itself.

    “all the tasks are in the github above. Our team don’t train the model”

SN64 24h −0.47%

Chutes

What happened
SN64 (Chutes) began rolling out additional inference models, Qwen3.8 Max and DeepSeek V4.1 Pro, in beta mode. These models run off-network under contract arrangements rather than being provided by network miners. Alongside this, the team ended subscription offerings, moving to PayGo-only access.
Community
A miner asked where the new models are running and whether regular miners would be able to run them later. A community member argued that subscriptions serve user acquisition and retention rather than profit and suggested reducing rather than removing the subsidy, while others debated the economics of subsidizing inference at a loss.
Official
The team said subscriptions are no longer feasible due to the mechanics of the new model offerings and current GPU/compute market dynamics, and stated they are no longer losing money on subscriptions.

More subnet community coverage every day in the Telegram channel →

Ecosystem debate

5 participants

Falling subnet registration price

Participants in the chat noted that the cost to register a new subnet on the network has been dropping quickly. Some projected the price could fall to a very low level within days, prompting comparisons with the separate cost of miner registration.

Positions

  • One community member projected the subnet registration price would soon be extremely low, within about a week.

    “The reg price will be 0.0001T after 8 days”

  • Another member suggested subnet registration could end up cheaper than miner registration.

    “it looks like the registration cost for a subnet will end up being lower than the registration cost for a miner”

  • A different member argued the current price level, around 200 TAO, was already reasonable and should not go lower.

    “lol, i think 200 tao is good”
Official response
There has been no official response from the team or Opentensor on this topic in the messages provided.
What to watch
Whether the subnet registration price continues to decline toward the levels projected in the chat, and whether the team comments on the trend.

Voices on X

  • @bittingthembits

    Grayscale, Bitwise, and REX-Osprey have all filed for U.S. TAO ETF products, while Grayscale's existing GTAO traded at a 22% premium to net asset value on Oct 23.

    View the original post on X ↗

  • @taoswap_org

    Taoswap launched a Bittensor-Ethereum bridge allowing USDC or ETH to be bridged into TAO or subnet tokens, and converted back, powered by Taostats.

    View the original post on X ↗

  • @chutes_ai

    Chutes is rolling out Qwen3.8 Max and DeepSeek V4.1 Pro in beta but says subscriptions are no longer feasible and existing ones will not renew.

    View the original post on X ↗

  • @b1m_ai

    Beam launched a Bittensor EVM integration putting data transfer and delivery receipts onchain for smart contracts and agentic workflows to read.

    View the original post on X ↗

  • @attelier_ai

    Attelier is rolling out video inference on Bittensor as part of a planned full stack for open video generation from decentralized GPU infrastructure.

    View the original post on X ↗

  • @TaoOutsider

    Subtensor merged a workflow that verifies a rebuilt Bittensor runtime matches the code recorded on-chain before it is published as a GitHub release.

    View the original post on X ↗

  • @Hippius_cloud

    Hippius launched a CDN product that caches files from S3-compatible buckets at nearby nodes, with HTTPS support and a spend cap per zone.

    View the original post on X ↗

  • @ShizzyUnchained

    SN92 will cut inference's share of the miner reward budget from 50% to 40% starting Oct 30, burning 10% of that budget.

    View the original post on X ↗

◆2026-10-27 · End of this edition

Protocol

2 PRs

New release · v486 Released

Subtensor v486 corrects how one smart-contract query is charged and makes runtime builds reproducible.

Runtime 486 is a small release. It changes how the chain accounts for the execution cost of one query that smart contracts can make into Subtensor, so that the charge better matches the work done. Staking, Emission and other network parameters do not change for TAO holders, Validators or Miners. The rest of the release is release tooling that makes runtime builds reproducible, so anyone can check a published runtime against the code running on chain.

Read the upgrade report →

#3289

release-486

What
Adds a missing weight/fee charge for a chain query that checks stake availability and bumps the network's internal version number to 486.
Why
The query previously could be called without being properly accounted for in transaction weight, which the fix corrects before the next network upgrade.
Who it affects
No direct effect on TAO holders, validators or miners; this is an internal node-software correction ahead of a version bump.

#3282

Make runtime builds reproducible and publish already-deployed runtimes

What
Changes how the node software is built so that compiling the same source code always produces an identical output, and adds a manual process to publish official release files for network versions that went live without following the normal release steps.
Why
Some versions of the software had reached the main network without a matching public release record, which broke an automated tool that watches for new releases and meant the published code could not be independently verified against what is actually running on the network.
Who it affects
No effect on TAO holders, validators or miners; this is a developer- and infrastructure-facing change to how releases are built and verified.

Recent editions

Past editions +
2026-11-012026-10-312026-10-302026-10-292026-10-282026-10-272026-10-262026-10-252026-10-242026-10-232026-10-222026-10-212026-10-202026-10-192026-10-182026-10-172026-10-162026-10-152026-10-142026-10-132026-10-122026-10-112026-10-102026-10-092026-10-082026-10-072026-10-062026-10-052026-10-042026-10-032026-10-022026-10-012026-09-302026-09-292026-09-282026-09-272026-09-262026-09-252026-09-242026-09-232026-09-222026-09-212026-09-202026-09-192026-09-182026-09-172026-09-162026-09-152026-09-142026-09-132026-09-122026-09-112026-09-102026-09-092026-09-082026-09-072026-09-06