Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Nov 21, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

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

THE DAILY BRIEFING

Bittensor Daily — November 21, 2026

The week's clearest protocol move landed in #Rao as a runtime upgrade rewired Gamma's spending mechanism, extending Nov 16's emission-control release into the deposit layer itself. Subnet-level scrutiny carried the rest of the day: SN55 (NIOME) disclosed deleted validator logs and prior credential theft after a miner found top-scoring UIDs missing from its records, while a community member raised sourcing questions about SN36 (Weir)'s stated location. #General logged its highest volume of the week, built mostly from routine onboarding and triage, including one member's end-to-end attempt to replace mainstream AI chat with a Bittensor-hosted model.

Today's highlights

  1. A runtime upgrade (v493) reworks Gamma deposit spending, shifting from a shared-spender model to per-hotkey signing and breaking existing integrations.

    Gamma deposits are now spent by their own hotkey: it signs a fee-free spend at a service listed on the subnet, the gamma is burned, and services credit customers from a per-service total instead of spending deposits. This breaks gamma integrations: the separate spender and GammaSpend proxy permission are removed, the mint, spend and set_service calls change arguments and call indices, and GammaRuntimeApi moves to version 2, so update the Python SDK before using gamma. A follow-up SDK note added SpendGamma now takes a service hotkey and an amount and is signed by the deposit's hotkey, and the new gamma_spent read returns what a hotkey has spent at a service. (1) This extends Nov 16's Release 491, which gave subnet owners emission-share and whitelisting controls, one step further into how Gamma deposits are actually spent. (2) The explicit breaking-change warning names removed calls, changed indices and a versioned API, a more detailed migration notice than Nov 18's release v492 coldkey-swap notes received. (3) No community reaction or question followed the release in the channel today.

  2. On SN55 (NIOME), a miner finds the top four scoring UIDs missing from the subnet's validator logs, which were then deleted, and the team discloses that stolen credentials could still compromise its services.

    (1) The miner reported that UIDs 251, 229, 159 and 137 received the highest incentives despite being absent from the wandb score logs, and the logs were deleted shortly after the discrepancy was raised. (2) The team said it does not directly manage the wandb service and that logs are deleted periodically, but also disclosed that credentials had previously been stolen, meaning some services including wandb could still be compromised. (3) The team stopped the validator and began burning emissions while investigating, separately announcing a pivot from synthetic genomic data generation toward AI model competitions; this follows the seed-matching analysis reported on Nov 17, which also went without a full official response.

  3. On SN36 (Weir), a community member alleges the team's old miner hotkeys show China-based IP activity inconsistent with its stated South Africa and Europe location.

    (1) The allegation follows Weir's beta launch of a GPU rental marketplace offering SSH access and a published FAQ on registration, provider setup and emission splits. (2) The team responded that it operates a mix of rented and idle GPU pools and reaffirmed it is a doxxed team based in South Africa and Europe. (3) No further verification or resolution of the IP-activity claim appears in the material.

  4. A #General member replaces a mainstream AI chat habit with a Bittensor-hosted model via Chutes, then hits repeated blocks from its safety classifier.

    Hi, i wanna use an alternative to the mainstream AI models for mundane things like helping me with searches and stuff. led to Okay, with some tokens in my wallet I'm now chatting with kimi-k2.6-tee. Thanks for the help, can't wait to see how it compares to what I've been using!, before the same member reported Sad to say that this is something I'm getting constantly from Chutes ai "This request can't be answered. Our safety classifier (Trishool Halo Guard) flagged it as unsafe, so the response was stopped. If you believe this is a mistake, please rephrase your message.", answered only with Please ask in the <#1320739778534047785> channel.. (1) This is a rare end-to-end onboarding narrative rather than a tokenomics or governance question, tying directly to Chutes' Alibaba Cloud partnership reported Nov 17. (2) The named safety layer, Trishool Halo Guard, links this consumer friction to Trishool's own Phase 4 roadmap announcement from Nov 17. (3) Neither the safety-classifier complaint nor the member's separate environmental-impact question received a substantive answer, both redirected to the subnet channel.

  5. Other: Carbon logs another round of intake outages, RedTeam disputes a deregistration exception, and a quantum-wallet rotation question goes unanswered.

    A #General member asked Is anyone using taoapp quantum resistant wallets? Which is the best to keep Tao secure for long term? Like ML-DSA-65 Rotation or fixed? without drawing any reply. (1) SN116 (Carbon) reported upload errors tied to a 32 KiB size limit and a build-machine overload affecting one miner's hotkeys, alongside three Motor 1.8 submissions returning FAILED_INFRA status with null scores. (2) SN61 (RedTeam) saw a dispute over whether a miner's submission should receive an exception from standard deregistration rules, with the team declining to allow manual database modifications. (3) #General logged 28 messages, its highest this week, driven by routine onboarding, scam warnings and social logistics rather than new tokenomics or governance substance.

Discord conversations worth following

#General

3 REPORTS

Notable

A member tries to replace mainstream AI chat with a Bittensor-hosted model and runs into a safety-classifier block and an unanswered environmental-impact question

Opening ·

A member asked Hi, i wanna use an alternative to the mainstream AI models for mundane things like helping me with searches and stuff., and was pointed to the Chutes channel: you can start by exploring <#1320739778534047785>.

Progress ·

After funding a wallet the member reported Yea, I checked out Chutes, thanks for mentioning it. I wanted to test it before commiting to it but I need tokens for it., then confirmed success: Okay, with some tokens in my wallet I'm now chatting with kimi-k2.6-tee. Thanks for the help, can't wait to see how it compares to what I've been using!.

Friction ·

The same member then asked Does anyone have information about the environmental impact of ai chats like chutes vs mainstream? and separately reported Sad to say that this is something I'm getting constantly from Chutes ai "This request can't be answered. Our safety classifier (Trishool Halo Guard) flagged it as unsafe, so the response was stopped. If you believe this is a mistake, please rephrase your message.".

Reply ·

Both follow-up questions were met with a single redirect: Please ask in the <#1320739778534047785> channel..

3 things to read from it

  1. This is a rare end-to-end narrative of a new user actually onboarding onto a subnet product rather than asking about tokenomics or governance, and it surfaces a concrete content-moderation friction point rather than a theoretical one.
  2. The named safety layer, Trishool Halo Guard, ties this consumer-facing friction directly to Trishool's own Phase 4 roadmap announcement reported on Nov 17.
  3. Neither the environmental-impact question nor the safety-classifier complaint received a substantive answer in the channel -- both were redirected to the subnet channel rather than addressed.

In the week’s context

  • Nov 17's edition reported Chutes (SN64) adding Alibaba Cloud as a second major cloud partnership within a week, giving today's end-user experience a direct line to that commercial push.

Where it stands

Partially resolved: the member successfully began using the product, but the environmental-impact question and the safety-classifier complaint both remain open, redirected rather than answered.

Notable

A member asks which quantum-resistant wallet rotation setting is best for long-term TAO security and gets no reply

Opening ·

A member asked Is anyone using taoapp quantum resistant wallets? Which is the best to keep Tao secure for long term? Like ML-DSA-65 Rotation or fixed? without drawing any response in the channel today.

2 things to read from it

  1. This is the third time this week a quantum-resistant wallet question has surfaced in these channels, after Nov 17's #General thread got a mechanism answer but no activation confirmation; today's narrower, more technical question about rotation settings went entirely unanswered.
  2. The lack of any reply, even a redirect, marks a step back from Nov 17, when at least a coldkey-swap mechanism was explained.

In the week’s context

  • Nov 17's edition recorded #General's first substantive engagement with quantum-resistant wallets, where a member's questions on activation status, staking continuity and migration timeline went unanswered despite a mechanism explanation being offered.

Where it stands

Unanswered: no community member or official voice responded to the rotation-setting question in the channel today.

Other

Other: in-person event thanks, a proxy-call error, hackathon test-TAO and registration questions, a pre-Finney archive question, and routine greetings filled out the day's highest #General volume of the week

Mixed ·

The channel opened with thanks for a meetup, Thanks to everyone that came today ❤️ let’s keep growing the in person events 💯, followed by a request for detail, At the first glance, it looks like they have done something really positive for Bittensor. Would like to understand exactly what they did. Any link for that?, answered with a social-media link: Thanks, I’ve started organising it, here is a link to the X for the socials: https://x.com/taosocials?s=11&t=JahwaMV9OfhgeZQ53jdChw. A proxy-call coding error drew only the standing scam warning, Anyone offering you a "registration script" in DM is a scammer (they're trying to exfiltrate your keys)., and a channel redirect. A hackathon participant asked Also, what's the recommended way for participants to get test TAO for the testnet deploy?, answered with an FAQ link and the same warning, Block all such requests. They're going to try to tell you that you need to connect your wallet to a scam website and give them your mnemonic phrase, so that they can regenerate your keys and steal your assets.. A separate member asked are the Transaction history and the blockstate for the pre-Finney era (before 20.03.2023) availble or achived somewhere ?, getting a pointer to You might want to ask in the Taostats Discord, though, which is linked in the "Third-Party Links" article. A newcomer's @-mention of another member prompted a joking suspicion, resolved with nothing sus xd he's new to bittensor, i invited him and Ah lol, quite often scammers just randomly @ people so thought it might have been that.

What to read from it

  1. Volume rose to 28 messages, the channel's highest point since Nov 17's spike to 34, but the increase was driven by routine triage, scam warnings and social logistics rather than new tokenomics or governance substance.

Where it stands

Closed; all items resolved through redirect, FAQ pointer, the standing scam script, or informal reassurance.

#Rao

2 REPORTS

Key

A runtime upgrade reworks how Gamma deposits are spent on-chain, shifting the mechanism from a shared spender model to per-hotkey signing.

Opening ·

An official release note announced the change directly: Gamma deposits are now spent by their own hotkey: it signs a fee-free spend at a service listed on the subnet, the gamma is burned, and services credit customers from a per-service total instead of spending deposits. The same message added that The funding coldkey keeps the balance and daily spend cap, and a hotkey swap moves the deposit.

Breaking change ·

The release note warned integrators directly: This breaks gamma integrations: the separate spender and GammaSpend proxy permission are removed, the mint, spend and set_service calls change arguments and call indices, and GammaRuntimeApi moves to version 2, so update the Python SDK before using gamma.

SDK detail ·

A follow-up message spelled out the SDK-level consequences: SpendGamma now takes a service hotkey and an amount and is signed by the deposit's hotkey, and the new gamma_spent read returns what a hotkey has spent at a service. It also confirmed removed tooling: SetGammaDepositSpender, btcli gamma set-spender, and the mint spender and service-hotkey options are removed, and a daily spend cap must be zero or at least 0.1 gamma.

Governance detail ·

The same release note added a separate sudo-level control: Sudo can also set each subnet's Yuma timelocked commit cap per hotkey, mechanism, and commit epoch; the default stays ten and zero pauses new commits while queued commitments remain eligible to reveal.

3 things to read from it

  1. This is the most detailed official release note logged in #Rao since Nov 16's gamma-emission-controls release, extending that subnet-owner-facing tooling one step further into how the deposits themselves are spent.
  2. The explicit breaking-change warning -- naming removed calls, changed call indices, and a versioned API -- gives integrators more specific migration guidance than Nov 18's release v492 notes on coldkey-swap liquidity handling.
  3. No community reaction, question, or complaint about the change appears in the material; the release stands alone in the channel today.

In the week’s context

  • Nov 16's Release 491 gave subnet owners gamma emission share and whitelisting controls, which today's spend-mechanism change extends further.
  • Nov 18's Release v492 reworked coldkey-swap liquidity handling at the protocol level; today's release is a second consecutive day-apart runtime change logged in #Rao with no community discussion following it.

Where it stands

Posted as an official announcement with no replies in the channel; whether gamma integrators have begun migrating to the new call structure is not addressed in the material.

Other

Other: a channel-routing question, a brief official check-in, and an emoji close out a day that rebounded from Nov 19's near-silence.

Routing ·

A member asked where is 128 channel?, answered with a channel pointer rather than discussion.

Check-in ·

An official account posted briefly, writing Here are your updates. with no further elaboration in the channel.

2 things to read from it

  1. Volume rose to 8 messages from Nov 19's single message, but the increase is almost entirely accounted for by the two release notes rather than any renewed community debate.
  2. Unlike the week's earlier #Rao threads on key custody, registration pricing, or governance alignment, no open question or disagreement carried over into today's remainder.

Where it stands

Closed; both items resolved without further reply.

Subnets

5 REPORTS

SN64 24h −0.42%

Chutes

What happened
SN64 (Chutes) limited how often a chute can be toggled off and on, allowing only one disable every 3 hours, after chutes were found thrashing GPUs and being blacklisted by miners. The team confirmed that buy/burn of the token has been paused since August following a company restructuring, and said it now prefers a 'gamma' auto-burn mechanism for tax and legal reasons, without giving a live date. Chutes added Alibaba's Qwen3.8 Max model through a partnership with Alibaba Cloud and is offering a 35% discount on DeepSeek V4.1 Flash until Nov 23. A shared post also pointed to deals between Chutes and Alibaba and Tencent.
Community
Community members criticized the removal of revenue transparency and the lack of value accrual to the token, with one member saying other subnets could surpass Chutes if nothing changes. Another member said questions about the buy/burn status have gone unanswered for months. A community member separately pointed to the Alibaba and Tencent deals as underappreciated news for Bittensor.
Official
The team said gamma is expected to go live "sometime soon" and that Parallax, a training initiative, is still being worked on, with a training run targeted before year-end following a roadmap posted by the project lead.

SN36 24h −5.25%

Weir

What happened
SN36 (Weir), a GPU rental marketplace, launched in beta, offering GPU rental with SSH access in about a minute. The team published a three-part FAQ covering registration, provider/miner setup, qualifying GPU models, and the emission split of 10% to rented GPUs and up to 90% to idle GPUs, with a 14-day eligibility period. A community member later alleged that the team's old miner hotkeys showed China-based IP activity and compute infrastructure inconsistent with the team's stated South Africa and Europe location.
Community
Members asked about mining eligibility, activation fees, and how the project compares to SN51 (Lium). Some questioned why five miner UIDs were earning rewards and pressed the team on its claimed doxxed status and the geographic origin of its infrastructure.
Official
The team confirmed it operates a mix of rented and idle GPU pools, stated the activation fee fluctuates with the token, and said it is a doxxed team based in South Africa and Europe.

In debate · Questions over Weir team's doxxed status and infrastructure location · 19 participants

  • A community member doubted the team's doxxed claim given use of a 'burner' Discord.

    “so you're doxxed? why are you using a burner discord then?”
  • The team maintained it is doxxed and based in South Africa and Europe.

    “South Africa and based in Europe.”
  • A member alleged the team's old mining hotkeys used China-based IPs and infrastructure inconsistent with its stated location.

    “your old miners, uid 61 and 65, both have use China Mobile IP in Jiaxing, Zhejiang, China”

SN55 24h +0.20%

NIOME

What happened
On SN55 (NIOME), a miner reported that the top four scoring UIDs (251, 229, 159, 137) were missing from the subnet's wandb score logs despite receiving the highest incentives. The wandb validator logs were deleted shortly after the discrepancy was raised. The team then stopped the validator, began burning emissions, and said it would investigate before making workflow changes. Separately, the team announced a pivot of Niome from synthetic genomic data generation toward AI model competitions, pausing competitions and continuing to burn emissions during the transition.
Community
Community members questioned why the logs were deleted immediately after the issue was flagged and asked whether the top miners were linked to the owner team. A moderator intervened to stop personal attacks in the channel and asked participants to focus on providing actionable information instead.
Official
The team said it does not directly manage the W&B service and that logs are deleted periodically, and disclosed that credentials had previously been stolen, meaning some services including W&B could still be compromised; it said security issues would be addressed before any workflow updates. The team later announced the shift to AI model competitions, stating competitions are paused and the validator stopped while emissions are burned during preparation.

In debate · SN55 top-miner scoring discrepancy and log deletion · 11 participants

  • Some community members concluded the top miners must be connected to the owner team.

    “The only explanation is that these top miners are part of owner-related teams.”
  • The team said the log deletion was a routine W&B practice rather than a response to the report.

    “We delete the validator run instance regularly in W&B.”
  • The team attributed the anomaly to a possible security breach rather than intentional action.

    “Some or all of our services including W&B, could be controlled by anyone if they have our creds or if the services are hacked.”

SN116 24h −6.25%

Carbon

What happened
SN116 (Carbon) miners reported submission upload errors tied to a 32 KiB size limit and piece-based uploads, which the team clarified involves compression and per-piece caps rather than a flat per-request limit. Separately, the team confirmed that builds on Nov 20 failed on both hotkeys for one miner because a build machine was overloaded during bandwidth testing, and said nothing was charged against daily allowance. Validating was also reported stopped and was fixed after an update, with additional compute added afterward to address slow evaluation speed. The team published a mainnet week 1 review for the Carbon physics AI subnet reporting 446 physics AI models built in 7 days, and three Motor 1.8 submissions later returned FAILED_INFRA status with null scores despite completed reconstruction.
Community
Miners asked for clarification on upload size limits and piece counting, and one miner suggested adding per-submission error logs to reduce back-and-forth with the team. Several community members also debated a leaderboard score, with one miner questioning whether it was fair and another defending the top position.
Official
The team clarified the 32 KiB compression-based limit, confirmed the build-machine overload was on their side with no charge to daily allowance, and added compute after validating issues were reported. It also said submission status tracking showing failure cause and daily spot usage is in progress, and is investigating the FAILED_INFRA cases on Motor 1.8 submissions.

SN61 24h +6.75%

RedTeam

What happened
SN61 (RedTeam) saw a dispute over whether a miner's submission, tied to UID201, should receive an exception from the subnet's standard deregistration rules; some community members questioned whether multiple miners pushing for the same exception were affiliated with each other. The team stated it does not allow manual database modifications to remove submissions from the comparison pool, citing fairness concerns. Separately, a miner reported that a BEX submission was marked invalid for violating security rules despite an earlier version passing, and another miner flagged that the ADA evaluation system was not running all 22 tests for each solution. The team also said a new challenge, delayed from earlier estimates, is still in progress and might be released this week.
Community
Miners argued over the fairness of granting an exception to UID201, with accusations exchanged about coordination between accounts and requests for wallet transparency. Other miners separately raised concerns about inconsistent BEX validation outcomes and instability in the ADA evaluation system.
Official
The team said manual database modifications to remove submissions are not permitted as a matter of policy, declined to share details on internal validation logic for the BEX case, and acknowledged the new challenge's release has been delayed from prior estimates.

More subnet community coverage every day in the Telegram channel →

Voices on X

  • @Rapido_ai

    SN120 (Affine) began sharing 10% of its emission with SN51 (Lium) via gamma, a flow now trackable live on taoswap.

  • @CryptoZPunisher

    A community member said Bittensor still has a communication problem, noting the SN110 (Green Compute) team's discussion with the team came only after the emission pause decision.

  • @taostats

    Taostats added cross-chain swaps, letting users move from Ethereum to Bittensor in one swap instead of using a bridge, exchange and withdrawal.

  • @chutes_ai

    SN64 (Chutes) added Qwen3.8 Max through a partnership with Alibaba Cloud, with the 0902 snapshot described as focused on multi-tool orchestration and autonomous development.

  • @markjeffrey

    A community member described gamma as Bittensor's new internal economy, letting AI products and agents mine together and pay each other on TAO rails.

  • @Hippius_cloud

    Hippius Cloud said SN91 (Cascade) cut storage costs by 66% and database costs by 33% after consolidating its stack with Hippius, with one publicly auditable record.

  • @404gen_

    404gen said its Titan 3D model, built by miners on Bittensor, now powers Atlas-based game asset generation for Parallel at about 50% lower cost than closed-source alternatives, per a Parallel executive.

  • @conjectures_io

    Conjectures said its miners ruled out degree-four nonlinear polynomial solutions to a 29-year-old minimal surface equation question, with the result verified in Lean.

◆2026-11-21 · End of this edition

Protocol

4 PRs

#3345 Breaking change

Release v493

What
This PR bundles release v493, which includes an adjustable per-subnet cap on Yuma timelocked commits, a new mechanism letting deposit hotkeys spend gamma fee-free at listed subnet services, and updated release tooling for announcements and Python publishing.
Why
The commit cap change gives subnets flexibility to throttle or pause commit submissions, while the gamma spend change lets deposits pay for services directly from their hotkey without a separate spender permission.
Who it affects
Validators and subnet operators gain control over commit limits per subnet; gamma holders and SDK integrators relying on gamma spending or the GammaSpend proxy must update to the new call structure, as it is a breaking change; ordinary TAO holders are not directly affected unless they use gamma services.

#3349

Add gamma hotkey spends to the v493 release notes

What
This PR corrects the v493 release notes and release page to include the gamma hotkey spending change, which had been left out despite being the release's main breaking change.
Why
The original notes only described the Yuma commit cap and release tooling, so gamma holders, service providers and SDK users were not warned about required updates.
Who it affects
This is a documentation fix; it affects how clearly gamma integrators and SDK users understand the v493 upgrade, but it does not itself change protocol behavior.

#3350

Publish stable Python and docs under the single mainnet approval

What
This PR changes the release process so that a single mainnet approval now also triggers the stable Python package publication and the production docs deployment.
Why
Previously these may have required separate approvals; consolidating them under one approval reduces manual steps in the release workflow.
Who it affects
This affects only the subtensor development and release team's internal workflow; it has no effect on validators, miners or TAO holders.

#3351

Fix v493 stable Python floor check and runtime announcements

What
This PR fixes two bugs in the v493 release train: a failure in the stable Python package version check caused by unpublished intermediate versions, and a failure in runtime upgrade announcements caused by an unexpected hex prefix in the wasm checksum.
Why
The version check could not resolve an installable package meeting the declared floor, and the announcement tooling rejected a correctly formatted checksum, both of which blocked the v493 release pipeline from completing.
Who it affects
This is an internal release-tooling fix affecting the development team only; it does not change subtensor's runtime behavior for validators, miners or TAO holders.

Recent editions

Past editions +