Daily digest of the Bittensor ecosystem

中文/EN

Daily edition
Nov 6, 2026

Bittensor Daily A 1T1B product
1T1B ↗
BITTENSOR DAILY

24H Market

01/10

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

THE DAILY BRIEFING

Bittensor Daily — November 6, 2026

Discord fell to its quietest combined volume of the week -- 10 messages in #General and 2 in #Rao -- leaving no new tokenomics or governance thread to follow the SN82 emissions and emission-decay debates that ran earlier. The day's substance moved instead to the protocol and the subnets: a breaking fee overhaul shipped and was partly reversed within days, SN108 (ChipForge) and SN114 (SOMA) each disclosed evaluation bugs that forced submissions back to a prior state, and the X feed carried a cross-subnet model result from Reliquary. No new ecosystem debate emerged in either Discord channel today.

Today's highlights

  1. Subtensor merges Release 488, reverting the staking, basket, proxy and EVM fee and gas-pricing changes bundled into Release 487 only days before.

    (1) Release 487, which also introduced the Gamma dollar-valued service credit and quantum-resistant wallet code, had bundled staking, basket, proxy and EVM fee and gas-pricing changes marked breaking. (2) Release 488 reverts those fee and gas-pricing changes to their prior values, removes an added pending-dividend surcharge, and restores 19 fee snapshots to their pre-release figures. (3) Both releases are marked breaking, meaning node operators and developers integrating with staking, basket, proxy or EVM functionality must update again so soon after the first change. No community reaction to either release appears in today's material.

  2. On SN108 (ChipForge), validator disagreement and EDA server crashes leave submissions stuck between queued and under-review, forcing a pause on new entries.

    (1) The team said the instability stems from two validators failing to agree on a score within tolerance, caused by machine timeouts and EDA server crashes, compounded by the complexity of evaluating full silicon designs rather than simpler AI outputs. (2) New per-round caps on how many mini and lite submissions get graded, meant to stabilize evaluation, have instead produced a growing backlog as submissions arrive faster than they can be processed. (3) The team paused new submissions to finish evaluator v2.0.5, confirming a fix for an OpenSTA memory-sizing bug affecting large MAC-heavy designs that a miner had reported; the subnet's token still rose 1.34% over 24 hours despite the disruption.

  3. SN114 (SOMA) reverts its CoT-13 competition after a benchmark proxy bug silently stripped reasoning data from Claude Code requests.

    (1) The team found that Claude Code requests in Anthropic format were being converted to OpenAI format before being forwarded upstream, causing loss of reasoning information during runs with anthropic/claude-haiku-5.5. (2) The bug forced a reversion of current competition uploads and a new baseline, with the team extending the upload window by one day to compensate. (3) The proxy was subsequently fixed to use the Anthropic request format exclusively; the subnet's token fell 7.09% over 24 hours.

  4. Reliquary releases Teutonic-I-Reliquary, a 9B model trained across four Bittensor subnets that beats Qwen3.5-9B on SWE-bench Verified.

    This is a full top to bottom model, pre-trained on Teutonic (3) and then post trained in a competitive environment to reach coding performance above Qwen's flag ship 9B model (1) @reliquary_ai said the model was post-trained on Bittensor using untrusted GPUs, reporting 43.6% versus Qwen3.5-9B's 35.1% on SWE-bench Verified. (2) A community member noted the model drew on SN81 (Reliquary), SN3 (Teutonic), SN51 (Lium) and SN4 (Targon), while Qwen still leads on three other benchmarks. (3) @const_reborn separately credited the compute behind the result to Lium and Targon, sourced competitively from anonymous operators.

  5. Elsewhere, #General and #Rao fall to their quietest combined volume of the week, with only routine moderation and a shared Novelty Search notice.

    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! Novelty is on tonight, starting Soon TM (1) Volume fell to 10 messages in #General and 2 in #Rao, among the lowest points of the week after Nov 3's 27-message moderation-heavy peak. (2) Neither channel saw the week's recurring tokenomics or governance threads -- the SN82 emissions dispute, the emission-decay proposal, or deregistration-price mechanics -- resurface. (3) The day's only cross-channel content was the shared Novelty Search livestream mention, also promoted on X by @opentensor.

Discord conversations worth following

#General

1 REPORT

Other

Other: a subnet-channel location question, a testnet-TAO request redirected to the FAQ, the standing scam warning, and a Novelty Search mention filled a quiet day

Opening ·

A member asked Hello, please let me know where I can check Subnet 103. I can't find the channel. Is there a problem on my end? and followed up no one know? before a moderator pointed them to the channel directly and added Use <#1179129432410173541> to get the subnet role.

Redirect ·

A separate member asked Could someone please send some testnet TAO to my address for testing a new GPU miner?, which a moderator answered by pointing to this post in the FAQ channel for information on how to get testnet tokens.

Reminder ·

The channel's regular scam reminder ran again, 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 flagging that The "Subtensor Wallet" and "Dsperse Wallet" are both SCAMS.

Close ·

A member noted Novelty is on tonight, starting Soon TM, echoed in #Rao's only message of the day, and the channel closed with a greeting, Good day everyone! Have a wonderful day 🫡.

3 things to read from it

  1. Volume held at 10 messages, close to Nov 4's 7, continuing the channel's post-Nov 3 falloff from that day's 27-message moderation-heavy peak; today's traffic was entirely routine support and redirects rather than ecosystem substance.
  2. The subnet-103 channel question mirrors the kind of manual-moderation gap the channel surfaced on Nov 3 with subnet 16's missing channel, though here the member received a direct link rather than an explanation of a deregistration-related cause.
  3. #Rao mirrored #General's thinness, logging only two messages today, both the same Novelty Search mention -- the week's tokenomics and governance threads (SN82 emissions, emission-decay, deregistration-price mechanics) did not resurface in either channel.

In the week’s context

  • Nov 3's #General similarly resolved a missing-channel question (subnet 16) through moderator explanation, though today's subnet-103 query was answered with a link rather than a cause.
  • The standing scam warning repeats language used across the week, including Oct 30's and Nov 2's double warnings, underscoring how much of #General's routine traffic remains anti-scam reminders.

Where it stands

All items closed through standard moderator redirect or reminder; nothing substantive was left open in the channel today.

#Rao

1 REPORT

Other

Other: a single message announcing the Novelty Search livestream was the only #Rao traffic of the day

Opening ·

A member posted word that Novelty is on tonight, starting Soon TM, the same notice relayed simultaneously in #General.

3 things to read from it

  1. This continues the channel's sharp falloff since Oct 31's 101-message peak -- Nov 4 already fell to 7 messages, and today's 2 messages mark the quietest day of the week by far.
  2. No tokenomics, governance or mechanism question resurfaced in #Rao today, leaving the week's SN82 emissions dispute and the emission-decay proposal without any further movement in this channel.
  3. The same Novelty Search event was promoted on X by @opentensor, giving the notice slightly more context than the bare Discord post offered.

In the week’s context

  • #Rao carried the week's governance debates on Oct 29-31 (shogting, emission-decay, SN82 emissions criteria) but has gone quiet since Nov 1, continuing through Nov 2-4 with no tokenomics or governance discussion.

Where it stands

Closed; a single event notice with no replies and nothing substantive to resolve.

Subnets

4 REPORTS

SN108 24h +1.34%

ChipForge

What happened
SN108 (ChipForge) experienced instability in its evaluation pipeline, with validator disagreements, timeouts, and EDA server crashes causing some submissions to be voided or left in "Under review" status. The team paused new submissions to complete evaluator v2.0.5, which addresses out-of-memory and compatibility bugs identified by miners, including an OpenSTA memory-sizing issue affecting large MAC-heavy designs. Previously voided submissions are expected to be re-evaluated once the new evaluator version is deployed.
Community
Miners raised questions about inconsistent evaluator versions across validators, the handling of voided submissions, and the indefinite block on new submissions. Some expressed concern about subnet stability, with one community member asking whether it would return to stable operation.
Official
The ChipForge team said the instability stems from the complexity of evaluating complete silicon designs rather than simpler AI outputs, and stated that getting the evaluation right takes priority over restoring the system quickly. The team confirmed the OpenSTA memory bug reported by a miner and said it is fixed in evaluator v2.0.5.

SN127 24h −12.00%

Zils

What happened
SN127 (Zils) launched its soup competition, in which miner-submitted models can earn spots in a private bank pool; models that do not earn a spot are not rewarded, and most emissions are currently burning due to low participation. The team reported a backlog of about 18 mini and 19 lite submitted models awaiting evaluation, since nightly grading is capped at 15 mini and 12 lite bodies per round. Two additional pools are planned to open later, giving miners more opportunities to earn a spot.
Community
Community members pressed the team on why evaluation runs only once per day and asked for faster or multiple daily grading rounds given the growing queue. Others asked about the per-round submission cap and the timing of the additional pools.
Official
The team said grading takes about four hours a day at current specs and is considering raising the per-round submission cap given the volume of commits. It also corrected earlier figures, noting the last two nights graded 33 new bodies combined across the mini and lite pools.

SN58 24h −4.68%

Attune

What happened
SN58 (Attune) rolled out an updated Attune Vector validator pipeline with private model submissions, gating models to authorized validators and automatically publishing weights for 'king' models. The team also launched Horizon, a new competition that lets users contribute robotics data via a phone app, with Robotensor confirming the app launch on X. A dedicated hotkey, UID 254, was set up as an owner-controlled vault receiving 70% of Horizon emissions and does not participate in Vector competitions.
Community
A miner asked whether UID 254's 67.7% allocation replaced the subnet owner's prior 70.2% share, noting the UID registered before the Horizon code shipped and has no model commitment or duel scoring. Other community members asked about dashboards, region and IP restrictions for Horizon, whether it is winner-take-all, and whether Vector's incentive pool should be weighted higher than Horizon's since Horizon requires no GPU cost.
Official
The team confirmed UID 254 is the owner-controlled Horizon vault hotkey, receiving the 70% Horizon emission share by design, and does not require model commitment or duel evaluation since it does not compete. The team also said Horizon focuses on data collection and real-world applications while Vector remains centered on robotics research, and that both are intended to generate value.

SN114 24h −7.09%

SOMA

What happened
The team identified a bug in the benchmark proxy used for CoT-13, where Claude Code requests in Anthropic format were being converted to OpenAI format before being forwarded upstream, causing loss of reasoning information during runs with anthropic/claude-haiku-5.5. This led to a reversion of current competition uploads and a new baseline. The team subsequently fixed the proxy to use the Anthropic request format exclusively and extended the upload window by one day.
Community
Some miners raised concerns about repeated rule changes during the competition and questioned the fairness of the scoring system, including how token usage versus solved runs affects scores. Other community members defended the token-efficiency scoring design and acknowledged that the proxy fix, while disruptive to prior tuning work, was necessary.
Official
The team stated it had not anticipated the bug, said it would run a new baseline and continue the current competition, and extended the upload window by one day. A separate update confirmed the benchmark proxy fix was deployed using the Anthropic request format exclusively.

More subnet community coverage every day in the Telegram channel →

Ecosystem debate

16 participants

ChipForge evaluation backlog and validator disagreement

Miners on ChipForge repeatedly reported submissions stuck between queued, evaluating, and under-review states, which the team says happens when two validators fail to agree on a score within tolerance due to machine timeouts or EDA server crashes. The team has since introduced per-round caps on how many submissions (mini/lite categories) get graded, which has produced a growing backlog as submissions continue to arrive faster than they can be evaluated.

Positions

  • Some community members argue the team should speed up or run evaluation more often given the size of the backlog.

    “I am not sure why you are not adjusting evaluation speed. Don't you want fast improvement?”
  • The team has defended prioritizing correctness of the evaluation pipeline over restoring speed quickly.

    “Our priority is getting it right, not rushing an unstable system back online.”
  • Another view holds that faster evaluation cycles would be counterproductive since leaderboard rankings would shift every epoch.

    “if evaluation is fast, the top miners will change every epoch. So we should accept the current situation.”
Official response
The team has acknowledged the instability publicly, explaining that ChipForge evaluates full silicon designs rather than simple model outputs, which makes the pipeline more complex than typical subnets, and that requiring two validators to agree within tolerance can lead to timeouts or under-review states. On the backlog, the team corrected a miner's figure, stating that the last two nights graded 33 new bodies (mini 1 + 15, lite 5 + 12), but that both rounds hit the per-round caps of 15 mini and 12 lite, leaving about 18 mini and 19 lite bodies waiting, and said they are modeling an increase to the caps.
What to watch
Whether the team raises the per-round mini/lite caps and whether that reduces the backlog of submissions waiting for evaluation.

Voices on X

  • @openroboto

    Openroboto signed its first commercial robotics data pilot, with Vulcan Shield Hungary, to train a factory-task robot model using real-world and simulated data in Békéscsaba.

  • @reliquary_ai

    Reliquary released Teutonic-I-Reliquary, a 9B model post-trained on Bittensor using untrusted GPUs, reporting 43.6% versus Qwen3.5-9B's 35.1% on SWE-bench Verified.

  • @ShizzyUnchained

    A community member reports that SN81 (Reliquary), SN3 (Teutonic), SN51 (Lium) and SN4 (Targon) contributed to the Teutonic-I-Reliquary model, noting the team's SWE-bench result and that Qwen leads on three other benchmarks.

  • @yanez__ai

    Yanez said it signed its largest client to date, a household-name social media company, with details withheld under NDA.

  • @robotensor_ai

    Robotensor launched Attune Horizon on SN58 (Attune), a phone-based app for collecting robot-training data, framing it against the scarcity of existing robot data.

  • @ShizzyUnchained

    A community member reports Subtensor release 487 merged to main with gamma service-payment code and quantum-resistant wallet code, while runtime remains at 486, and flags unverified payouts on SN58 (Attune) and a pending SN38 (ChronoLLM) model preview.

  • @chutes_ai

    Chutes' weekly update for Oct 21 to Nov 5 covers seven new catalog models, a payment-system change, a SayGM listing, and other platform updates.

  • @YumaGroup

    Yuma Group reported its Total Market Fund fell 20.4% in USD in October versus TAO's 20.9% decline, gained 0.6% in TAO terms, and remains up 4.4% since launch.

◆2026-11-06 · End of this edition

Protocol

3 PRs

New release · v488 Released

Subtensor v488: Gamma credits, quantum-resistant wallets, Lock proxy and fee restoration

This release bundles the v487 feature set with v488 follow-up fixes. It adds Gamma, dollar-valued prepaid credits for subnet services, and quantum-resistant wallets with key rotation. It also adds a Lock proxy type, lets Staking proxies claim root dividends, and allows nonzero burn and recycle emission shares. v488 restores the fees and gas requirements that v487 had raised for some calls and fixes root claims for full baskets.

Read the upgrade report →

#3294 Breaking change

Release 487 — Gamma + Quantum-Resistant Wallets

What
Introduces Gamma, a prepaid dollar-valued credit system that subnets can use to charge for services, funded by burning subnet alpha, plus other changes bundled into Release 487 that the team has flagged as breaking.
Why
The change gives subnets a mechanism to bill for services directly on-chain and ties that billing to burning subnet alpha rather than ordinary staking.
Who it affects
Subnet owners and validators with a permit can publish services and set up billing; TAO holders who fund a Gamma deposit convert TAO or alpha into non-transferable, non-redeemable subnet-specific credit; miners are not directly affected by this change.

#3304 Breaking change

Release 488

What
Reverts the staking, basket, proxy and EVM fee and gas pricing changes introduced in Release 487 back to their prior values, removes an added pending-dividend surcharge, and restores 19 fee snapshots to their pre-release figures.
Why
The adjustments made in Release 487 to fees and gas pricing were rolled back, and a separate budget was added so the full 129-row basket's root claim can still process pending dividends.
Who it affects
Validators, miners and TAO holders who pay fees for staking, basket, proxy or EVM operations will see pricing return to pre-Release 487 levels; this is a corrective change rather than a new feature.

#3307

fix: add support for signing native extrinsics with ECDSA keys

What
Fixes a bug in the Python SDK where ECDSA signatures were incorrectly truncated because the signing code assumed all signatures were a fixed shorter length used by other key types.
Why
The SDK now distinguishes signature length by key type so ECDSA signatures are handled correctly instead of being cut short.
Who it affects
This affects developers building on the Python SDK who sign extrinsics with ECDSA keys; it has no direct effect on TAO holders, validators or miners outside of tooling that relies on the SDK.

Recent editions

Past editions +