October 20, 2026 · Published daily
Subnet News — October 20, 2026
eTAO is live, with nine executed allocation changes documented, including SN103’s old slot before re-registration. SN120 Affine opens a video-model dashboard, SN28 SayGM launches app sign-in, SN114 SOMA adds Claude Code, SN71 Leadpoet expands into outreach, and SN65 TPN publishes model 006. Also: SN66 Conjectures publicizes an earlier certified result, operational updates, SN80 OpenRoboto’s teaser and the new Doc interview. Late update: SN16 Zero Hour and SN47 GPUForge shift more alpha rewards toward miners.
- Coverage window
- Oct 19, 2026, 10:40 AM – Oct 20, 2026, 12:35 PM ET
- Subnets in coverage roster
- 128
- Subnets with updates
- 33
Biggest stories first
Today's ranked news
eTAO reaches mainnet: subnets can choose their alpha reward split
Finalized Finney runtime 481 is active. Owners can adjust owner, validator and miner allocations rather than use the universal 18/41/41 default. Root receives its live baseline from the validator bucket, capped by that bucket; alpha validators share the remainder. This changes distribution, not the subnet’s network-emission allocation or a guaranteed APY. Runtime 481 removed the earlier per-subnet TAO-weight approach: consensus still uses the global TaoWeight. Our eTAO page separates verified transactions from snapshot observations.
SN8 Vanta removes its owner allocation and reports trader payouts
At October 20 00:14 UTC, a successful transaction changed SN8 Vanta from approximately 18/41/41 to 0/50/50 for owner/validators/miners. Its founder describes removing owner emissions as a maturation step, not a prohibition on team participation in incentives. Separately, SN8 Vanta reports $930,128 across 339 rewards to 122 traders, and over 5,000 registered traders since February. Those totals were not independently reconciled against every ledger entry; the split change itself is verified on-chain.
SN16 Zero Hour raises miner allocation to 61.5% through eTAO
SN16 Zero Hour executed an eTAO allocation change on October 20 at 16:37:12 UTC, in finalized transaction 9325968-13. The owner allocation remains approximately 18%; validators fall from approximately 41% to 20.5%, while miners rise from approximately 41% to 61.5%. Successful execution and the exact five-slot storage were checked immediately before and after the transaction; burn and recycle remain zero. This redirects a larger share of alpha rewards toward miners. Root receives its live baseline first, capped by the validator bucket, and alpha validators share the remainder. Historical yield can still include the previous split, so this change does not establish a specific new APR or APY.
SN28 SayGM launches app sign-in with permissions and budgets
SN28 SayGM says apps can now use OAuth 2.0 with PKCE so users bring their existing balance instead of pasting API keys. Its October 20 announcement describes separate model, usage and profile permissions, per-app monthly budgets and immediate revocation. Developers can attach affiliate referrals under the programme’s terms. The change reduces integration friction; it does not guarantee app safety or savings. The miner repository separately publishes 0.5.5 following connection-handling fixes.
SN41 Almanac raises miners to 50% and updates its market-data integration
The October 19 19:50:48 UTC transaction set SN41 Almanac to 10% owner, 40% validators and 50% miners. Related code updates default miner alpha and restores market weight to 95% for the larger miner pool; a separate change migrates its Polymarket script to Data API v2. These are transaction and code changes, not proof of better trading performance. Its roughly 18.4% 24-hour alpha/TAO rise is a timestamped observation, not a demonstrated causal effect.
SN47 GPUForge raises miner allocation to 69.4% through eTAO
SN47 GPUForge executed an eTAO allocation change on October 20 at 16:27:12 UTC, in finalized transaction 9325918-26. The owner allocation remains approximately 18%; validators fall from approximately 41% to 12.6%, while miners rise from approximately 41% to 69.4%. Successful execution and the exact five-slot storage were checked immediately before and after the transaction; burn and recycle remain zero. This redirects a larger share of alpha rewards toward miners. Root receives its live baseline first, capped by the validator bucket, and alpha validators share the remainder. Historical yield can still include the previous split, so this change does not establish a specific new APR or APY.
SN65 TPN publishes its 006 compressed model
SN65 TPN says model 006 is live on Hugging Face, reporting 74.93% on fluid_knowledge at 7.71GB versus 66.57% and 17GB for its target. The linked bboldpilot/tpn-006-sub-g artifact is publicly accessible. Those are team-reported benchmark and size figures, not independently rerun measurements or universal quality claims. Revised KL-divergence, quantization and broader architecture support remain TPN-007 plans; the October 15 competition opening is earlier news.
SN66 Conjectures publicizes an October-certified dimension-eight result
SN66 Conjectures’ October 20 post highlights a result credited to JenW1N. Its public record says Lean verification passed, review was approved and the reward paid; certification is dated October 14 and verification October 8. The recorded claim is that every scalar real polynomial minimal graph in eight variables is affine, across all polynomial degrees. The site discloses that a second independent kernel was not run. This is today’s announcement of an earlier certification, not a proof completed today or an independent literature audit.
SN71 Leadpoet expands from prospect research into automated outreach
SN71 Leadpoet’s October 20 launch thread and live website describe finding buying signals, researching prospects, writing messages and following up through email and LinkedIn. The sampled demonstration uses contract-signing software; users review messages and choose a schedule. The team names Dropbox, OnePatch and Muttdata as users, but those relationships and revenue were not independently confirmed with the customers. The concrete move is a fuller sales workflow rather than lead identification alone.
SN92 REAX gains an Antseed channel while Eval v3 rolls out
Decision Models identifies itself as powered by SN92 REAX and announced an Antseed listing on October 20, including an EU tier and automatic router. The linked reaxlabs/auto-model-router repository supports a concrete tooling update. Separately, the channel digest dates Eval v3 activation to October 19 at 18:08 UTC and schedules the engine track for October 20 at 18:00 UTC, with a possible next-day fallback. The approximately 664% seven-day alpha/TAO rise is not proof of successful scoring or adoption.
SN114 SOMA adds Claude Code support
SN114 SOMA’s October 20 announcement adds Claude Code alongside existing Codex and Copilot integrations, expanding the coding tools that can access the service. That is a product integration, not a new model benchmark. Separately, the dated channel digest reports eight round-149 submissions flagged for disallowed prompt injection and a retired evaluation model preventing some reruns. Those enforcement details remain attributed to the digest rather than independently adjudicated findings.
SN120 Affine opens its video-model dashboard after Const’s first-release post
Const’s October 20 post links video.affine.io as a first model release and includes a five-second generated sample showing a worker handling rope on a boat. The live product identifies SN120 Affine, mechanism 1, with champion Veda-Sparse/Minimax-H3-T2VA-Veda-8NFE-600Step-Preview, duel scores and on-chain submission history. This is a concrete new video-model surface, not independent proof of benchmark leadership or the generation-cost estimates in replies.
SN1 Apex moves to a tougher competition iteration
The October 20 channel digest reports v0.1.5 activation at round 4 and a one-day reveal period for new competitions. A public account reports the NIST-Lite board nearly maxed out at 639/640, giving context to harder evaluation discussions. These are attributed operational reports rather than independently replayed results. Expectations for v0.1.6 are not a completed release.
SN7 Allways reports reopening after its halt
The October 20 team-channel digest says SN7 Allways resumed with 3.7.0, reopened swaps, reservations and deposits, and planned a return to zero burn. The repository contains new payout-delivery and gas-floor safeguards. Full recovery and the burn transition were not independently verified on-chain in this pass; miners and validators should follow version-specific team instructions.
SN9 IOTA reports faster decentralized-training merges; Const endorses it
SN9 IOTA’s October 19 team post, quoted by Const, says a tree-based merge graph makes merges about five times faster and training almost 20% cheaper, reducing swarm waiting time. These are reported measurements, not independently reproduced benchmarks. Const’s description of the team as the grand masters of decentralized training is commentary, not separate technical verification.
SN22 Desearch publishes an MCP agent example and updates embed checks
SN22 Desearch’s October 20 post demonstrates an agent using hosted MCP to discover tools, read pages and produce sourced answers with a cost log; the team says the default scripted loop needs no LLM. Code publishes 0.0.238 and longer reference-vector waits during server failover. A channel statement forecasting embed mining in one hour is a schedule claim rather than independently confirmed activation.
SN35 Unknown changes its allocation to 33/30/37
A successful October 19 transaction set SN35 Unknown to approximately 33% owner, 30% validators and 37% miners, replacing the prior 18/41/41 split. Root’s baseline is paid from the validator bucket, subject to its cap; alpha validators share the remainder. The chain establishes execution, not the owner’s motive, a sale or future yield. A substantial 24-hour alpha/TAO decline is also visible in the market snapshot, but timing alone does not prove the allocation caused the move.
SN36 Epago changes its allocation to 33/30/37
A successful October 19 transaction set SN36 Epago to approximately 33% owner, 30% validators and 37% miners, replacing the prior 18/41/41 split. Root’s baseline is paid from the validator bucket, subject to its cap; alpha validators share the remainder. The chain establishes execution, not the owner’s motive, a sale or future yield. A substantial 24-hour alpha/TAO decline is also visible in the market snapshot, but timing alone does not prove the allocation caused the move.
SN44 Score reports 98% fallen-person detection and publishes scoring fixes
SN44 Score reports fallen-person detection rising from a 43.6% baseline to 98% on crowded-street footage, above its 90% target, with harder footage planned next. This is task-specific team evidence, not independently reproduced real-world accuracy. Public code also contains archive/backfill and tie-break fixes matching the channel digest. Individual copying allegations are not treated as adjudicated facts; beverage-detection problems remain unresolved in the inspected evidence.
SN50 Synth code restores a 22% owner-UID weight allocation
An October 19 commit labelled “burn 22% of miner emission” assigns the owner UID 22% of the validator’s normalized miner weights and clears stale scores before the complete allocation. The code and tests establish the weighting rule, not independently verified mainnet activation or proof that the resulting tokens were destroyed. This is distinct from changing the five-slot eTAO owner/validator/miner vector; the current eTAO snapshot for SN50 Synth remains the default. Miners should distinguish a validator’s weight allocation from network-level emission shares.
SN55 NIOME reports a validator rollback; v4 remains on testnet
The October 20 team-channel digest says compression/backup changes caused failed uploads and inconsistent benchmarks, prompting a rollback. It reports v4 on testnet, with mainnet dependent on verification and a stable validator rerun expected. This is a developing incident and staged release, not a verified mainnet fix or all-clear. The earlier migration plan is not realized capital.
SN70 Consted 2.0 changes its allocation to 33/33/34
A successful October 19 transaction set SN70 Consted 2.0 to approximately 33% owner, 33% validators and 34% miners, replacing the prior 18/41/41 split. Root’s baseline is paid from the validator bucket, subject to its cap; alpha validators share the remainder. The chain establishes execution, not the owner’s motive, a sale or future yield. A substantial 24-hour alpha/TAO decline is also visible in the market snapshot, but timing alone does not prove the allocation caused the move.
SN78 UMI’s AMA recap moves attention to C6 intake
SN78 UMI’s October 19 recap confirms the previously covered C5 rewards and says C6 uses direct R2 uploads for concurrent miner submissions. Intake depends on fixes and is intended to last at least about 12 hours; the post forecasts an October 20 start, not independently confirmed as completed here. Trupan’s C5 model becomes the baseline; possible LLM scoring is a C7 idea. The roughly 15.5% 24-hour alpha/TAO decline does not establish an operating cause.
SN91 Cascade adds Timestral to Ephemeris
SN91 Cascade says Mistral’s 209M Timestral time-series model is live on Ephemeris as of October 19, with access through its MCP for compatible agent tools. This adds a forecasting option to the product. The team’s SOTA wording is not independent proof of outperformance on every task, and this is distinct from the earlier paper-trading result.
SN97 Ingot tightens model-validity checks and score display
Public changes in SN97 Ingot reject challengers that remove protected unquantized modules, improve skip-list and size checks, and display non-finite scores more safely. The digest reports A7 iterations and a duplicate filter focused on near-identical reuploads. These are evaluation-rule and reliability updates, not a newly benchmarked model; eligibility remains distinct from merely appearing on a leaderboard.
SN103’s old slot changed rewards before removal and re-registration
The old SN103 slot moved to 33% owner, 25% validators and 42% miners at block 9,319,453 on October 19. Historical storage shows that configuration cleared at block 9,321,588, October 20 02:01:12 UTC, with SubnetIdentityRemoved. NetworkAdded followed one block later, at 02:01:24 UTC; registration height and owner coldkey changed. The re-registered slot now has the default split and no confirmed current name. This establishes slot turnover, not a negotiated sale, buyer identity or price. Its extreme market return should not be attributed uncritically to the prior project.
SN105 Beam changes epoch-tier weights in public code
SN105 Beam commits multiplier-based epoch-tier weights. The digest separately describes PRISM self-dealing screening and a recovery-task distribution fix as activated October 19. The code is readable, while every production effect was not exhaustively verified. A published fix does not prove all historical transfer disputes are resolved.
SN118 Ditto details referral and business-partner programmes
SN118 Ditto’s October 20 recap describes programmes introduced at the September 28–29 Exploit Summit: usage-qualified referrals earn credits, and business introductions earn commission when customers pay. The team describes eight-week referral seasons and a top-three alpha prize pool. This is detail about an earlier event, not new verified revenue. The channel digest separately reports ten I2-based rejections reversed after a policy misapplication; that ruling remains attributed.
SN122 CookingTAO raises miners to 64%; holder credits remain planned
The successful October 19 transaction set SN122 CookingTAO to approximately 18% owner, 18% validators and 64% miners. The team attributes this to inexpensive validation and a desire to reward miners; the rationale is attributed while the allocation is on-chain. Its October 20 post separately promises monthly platform credits for alpha holders and one-click mining, with production release still to come. A later question about reducing the owner share is a proposal, not another executed change or guaranteed APY increase.
SN124 Swarm confirms its October 15 Andorra office opening
SN124 Swarm’s October 20 post says Andorra’s Secretary of State for Business, Economic Diversification and Innovation attended the October 15 office opening alongside investors, clients and industry leaders. It names EverBlueGreen and reports over 40 attendees. This is the team’s confirmation of an earlier event, not contracts signed today. Secondary/community claims about a EUR370,000 raise, CTO appointment and Q1-2027 drone test were not independently confirmed in the inspected primary material.
Doc interview joins the library; today’s Pareton stream remains upcoming
Shizzy’s interview with Invest With Doc, “The BEST Bittensor Subnets To Buy Before The Next Leg Up! Ft. Shizzy”, is added as a guest appearance. The latest channel episode is “Bittensor TAO vs AI Tyranny: Who Controls the Future?”, published October 19. The SN10 Pareton interview is scheduled October 20 at 17:00 UTC / 1 PM Eastern; SN58 Attune follows October 21 at 16:00 UTC / noon Eastern. Scheduled streams are not labelled live before their start.
SN54 Yanez postpones its October 21 Pulse workshop
SN54 Yanez’s October 19 primary post moves the Pulse workshop to November, with the replacement date still to be announced. Existing registrants are directed to the team’s support group. The original event is no longer scheduled for this Wednesday, and no confirmed November session time is supplied.
SN80 OpenRoboto’s next-chain teaser draws Shizzy’s Solana speculation
SN80 OpenRoboto’s October 20 post recaps its existing Base and Robinhood Chain presence and asks what comes next; the image repeats those milestones. Shizzy quoted the post and asked whether Solana could be next. That is owner commentary, not a team announcement of a Solana integration, launch date or new deployment. Both originals are linked so readers can distinguish the teaser from the speculation.
Const highlights SN88’s Root strategy; comparative returns remain a team claim
Const’s October 20 quote praises SN88 Investing88’s use of Root for recursive self-improvement. The quoted chart argues its basket outperforms other validators using lifetime observations. This is commentary on a team-selected comparison, not an independently normalized ranking or staking recommendation. Window, fund life, fees and realizable holdings must match before comparing returns.
SN107 Minos previews its app; Montreal remains scheduled
SN107 Minos’ October 19 post says to bring a model and prepare an agent, with a launch date coming soon. The October 20 digest expects a date later this week; neither proves a completed app launch. Yesterday’s ASHG/Montreal coverage remains upcoming October 22–23, with exact session titles, room and times still unverified here. No commercial OpenAI investment or partnership is inferred.
Audit trail
Coverage notes
Searched
- 128 current subnet profiles and dated channel digests; current/legacy repositories
- 214 public team/partner account samples and 269 original post pages
- Signed-in @ShizzyUnchained Grok discovery, corrected-cutoff top-five and omission checks; direct Const Posts/Replies
- Owner Bookmarks and History Likes, with saved lead dispositions
- Finalized owner/registration/presence comparisons and nine verified eTAO changes, including historical SN103
- Full-roster market outliers, emissions, Moments and seven prior editions
- YouTube Videos/Live tabs, watch-player metadata and guest interview verification
Public-data gaps
- Public sampling and qualitative Grok rankings are not an exhaustive X census. Private Discord originals and some attached media remain inaccessible or partially inspected; digest-only statements stay attributed.
- Archive RPC denies historical state_queryStorage (RPC4003). Specific events and endpoint changes were verified, but complete replay of all intermediate changes is not claimed. No independently confirmed subnet-sale consideration was established.
- Team benchmark, customer, payout, encryption and cost claims were not independently audited or reproduced. Conjectures dates come from its result record, without an independent full mathematical/literature review.
- TaoStats-backed server routes were sampled without exposing keys; credit usage and full direct event reconciliation remain unavailable. Initial root and candle requests returned503 but both recovered to200 on recheck. No new fallback integration or complete historical backfill is claimed.
- Bookmarks and Likes are not a complete viewing history. SN80 Solana remains speculation; exact SN107 conference sessions remain unverified.
- Targeted eTAO supplement checked through finalized block 9326441 at 18:12 UTC: SN16 and SN47 transactions added after the main research cutoff. Other-source coverageEnd remains 16:35 UTC; no complete new all-source scan or intermediate chain-event replay is claimed.