October 1, 2026 · Published daily
Subnet News — October 1, 2026
SN22 Desearch builds its own web index, SN28 SayGM powers SN105 Beam AI, and SN14 Cacheon names SN120 Affine as its first customer. SN50 Synth adds volatility scoring; SN81 Reliquary reports four training jobs. SN78 UMI intake is open but rewards remain inactive, and SN120 Affine documents its temporary reward transition. Plus SN38 ChronoLLM’s attestation fix, SN104 TAOstatus testnet and today’s livestreams.
- Coverage window
- Sep 30, 2026, 10:52 AM – Oct 1, 2026, 11:10 AM ET
- Subnets in coverage roster
- 128
- Subnets with updates
- 30
Biggest stories first
Today's ranked news
SN14 Cacheon identifies SN120 Affine as its first customer
SN14 Cacheon’s Cameron Fairchild identifies SN120 Affine as its first customer in a September 30 post. The official SN14 Cacheon account had already shown the collaboration in Const’s Exploit keynote on September 28. The new first-customer statement makes that relationship more concrete, but does not disclose commercial terms, measured savings or revenue.
SN22 Desearch 2.0 starts building its own web index
SN22 Desearch’s September 30 announcement and new 2.0 code describe a competitive web index for AI. Crawl miners fetch batches of pages, while validators re-fetch samples and credit verified work. The public task API returned October 1 crawl verdicts during this scan, supporting actual operation of the crawl track. GPU embeddings are built but await the SN22 Desearch embedding model; retrieval and ranking fine-tuning are later programs. The existing search API should not be confused with every new track being live.
SN28 SayGM powers AI prompts inside SN105 Beam Studio
SN28 SayGM confirms it is a launch partner for SN105 Beam Studio and says every Beam AI prompt, from building workflows to investigating failed runs, routes through its service. SN105 Beam acknowledges the announcement in a direct reply. This adds a named subnet customer to the inference story after Studio’s launch. Usage volumes, contract terms and revenue were not disclosed in these sources.
SN50 Synth adds live volatility scoring to its one-hour competition
SN50 Synth announces Volatility CRPS is live in the Crypto 1h competition. It scores the volatility implied by miners’ 1,000 simulated price paths against realized volatility across the full hour, four 15-minute blocks and twelve five-minute blocks. The point is to reward a better forecast of the risk along a path, rather than only where the price ends. The team’s announcement establishes the scoring change; it does not prove profitable trading performance.
SN78 UMI opens C5 intake; rewards remain inactive
SN78 UMI’s public competition API confirms open intake, available admission capacity and ready endpoint/model evaluation tracks. At this scan it reported 165 accepted submissions under the current policy, which allocates 50% to endpoint work and 50% to models. Crucially, the API still says rewards_active: false and chain_submission_authorized: false, in intake_no_weight mode. The C5 launch code and miner guide are progress; they do not establish resumed reward payments.
SN81 Reliquary reports four SFT jobs with separate incentive pools
SN81 Reliquary’s October 1 team-attributed brief reports four supervised fine-tuning jobs: math, code, logic and instruction following. Each has an incentive pool tied to verified tokens, and miners are told to update main and select a job with --job-id. Public code adds hot job loading, job-status routes and remote audit executors. This is distinct from yesterday’s customer-pilot invitation. Live job deployment and token payouts remain team-reported rather than independently executed here.
SN104 TAOstatus runs its security track on testnet
SN104 TAOstatus announces a second mechanism for security-audit agents alongside its LLM-key track, using fresh Docker targets to test vulnerability discovery and reproduction. The team explicitly says it is live on testnet. Mainnet is targeted for Friday, October 2, after successful testing; it has not been announced complete. Public validator code supports the dual-mechanism testnet setup and evaluation safeguards.
SN120 Affine publishes temporary equal-registration reward rules
SN120 Affine’s operational record describes a September 30 transition away from an owner-only burn job toward equal 1/N weights for eligible newly activated model registrations. The cutoff is Finney block 9181759, September 30 at 15:54:34 UTC. Eligibility requires a verified uploaded model, a qualifying activation/reveal and a currently registered hotkey; old activations and registration alone do not qualify. The submission window ends October 2 at 15:54:34 UTC, or 11:54 AM ET. The rules explicitly say no eligible models means no new weight submission, and expiry does not automatically restore old payouts. This is the documented policy and operational report; finalized current payouts were not independently reconciled.
Kusanagi reports exploratory Bittensor discussions at MIT Media Lab
Kusanagi says its team visited MIT Media Lab on September 30 to present Bittensor and met leadership and researchers for initial discussions about a possible relationship involving the wider ecosystem through OpenTensor. This is Kusanagi’s account of exploratory talks. A signed partnership, MIT endorsement or deployed research program was not confirmed.
SN4 Targon advertises on-demand NVIDIA B300 capacity
After yesterday’s cutoff, SN4 Targon says NVIDIA B300 instances are available on demand and links its inventory. This adds a specific hardware availability update to the already-covered Bare Metal and Sandboxes launches. Inventory can change, and no reservation or purchase was executed during this scan.
SN5 Hone widens task offers and avoids grading identical answers repeatedly
SN5 Hone merges a change offering tasks to 128 miners instead of 32, grading byte-identical submissions only once, and sizing grading concurrency from the validator host’s CPU, RAM and free disk. This is a concrete throughput and resource-management update. The implementation is public; fleet-wide deployment and measured speed gains were not independently established.
SN6 Numinous reports reasoning scoring at 10% of emissions
Today’s team-attributed brief says reasoning scoring is live and represents 10% of subnet emissions, while the frontend has not yet caught up. The brief also describes concentration among the leading miners, but its wording about individual percentages is ambiguous, so no per-miner share is asserted here. The allocation is a team-reported change; finalized rewards were not independently checked.
SN13 Data Universe fixes default miner startup and a slow validation scan
SN13 Data Universe removes a stale YouTube scraper configuration that caused default miner startup validation to fail. A separate change limits a job-window scan to locally cached files so it does not add a remote download scan to validation. Documentation now reflects current on-demand reward checks. These are verified code fixes, rather than a claim that every operator has updated.
SN15 ORO outlines a lower qualifying cap and refreshed tasks
Today’s team-attributed brief outlines a $2 qualifying inference cap, with the race limit staying at $5, a 62-model allowlist and a refreshed qualifying pack. It also promises a local test pack and failure-category feedback. The announcement mixes immediate and upcoming work, so the complete bundle is not presented as already deployed.
SN38 ChronoLLM reports a fix for miner attestation rejections
Today’s team-attributed brief identifies a backend TEE attestation check as the cause of miner rejections and reports support for the new dstack attestation format. The same brief contains miner reports of successful self-tests afterward. This is an operational follow-up to the live time-aware model announcement, not a new model launch. Recovery remains developing because a fresh miner admission and reward outcome were not independently verified.
SN44 Score reports progress on petrol-station detection
SN44 Score reports its DetectPetrolStation task score rose from a 46.2 baseline to 89.1, with 90 as the next target. The team identifies pump and hose detection as remaining weak points. These are task-specific team-reported scores, not a general accuracy measure. Separately, today’s channel brief says the Person element has been down for hours and a fix is being worked on; the petrol result does not establish that recovery.
SN51 lium.io prevents repeated executor listings from being counted twice
SN51 lium.io merges deduplication so repeated listings of an executor within a miner’s response are validated and counted once. Another update removes an Inspector enforcement flag that the commit says was never enabled in production; findings remain recorded without affecting scores. Separately, new spot-revenue work retains default-off gates and is not presented here as activated rewards.
SN61 RedTeam buybacks still depend on Kraken onboarding
Today’s team-attributed brief says a final Kraken KYC document is still required before buybacks resume, with the team intending to backdate funds. The cited wallet balance is existing locked alpha, not proof of a new purchase. The headline of the secondary brief should not be read as completed onboarding or a confirmed buyback restart.
SN66 Conjectures reports reopened submissions after maintenance
Today’s team-attributed brief reports maintenance finished and submissions reopened. It also describes an admission stall lasting roughly five and a half hours, followed by community reports of resumed evaluations. The reopening is attributed to the channel summary, and a fresh completed evaluation was not independently submitted here. This follows the previously covered DEFLATE launch.
SN71 Leadpoet publishes a recovery path for provider-scoring holds
SN71 Leadpoet publishes an October 1 migration to release a provider-scoring hold and retry exact scores. The accompanying changes preserve accepted outputs rather than silently replacing the work when a scoring provider fails. This is a recovery implementation, not independent confirmation that the migration has run or all held evaluations have cleared.
SN78 UMI discusses distribution and integration with Convo
SN78 UMI says its relationship with Convo is taking shape around combining AI overflow with trusted human interpretation and expanding distribution. The September 30 post describes discussions and work in progress, with more to follow. It is relevant commercial direction for the sign-language project, but these sources do not confirm a completed contract or a deployed Convo integration.
SN91 Cascade revises its champion-margin rollout rules
SN91 Cascade publishes a lower champion challenge-margin schedule, falling from 0.4 to 0.2 over 18,000 blocks of tenure. Its latest correction requires the change to begin at a fresh era after the readiness threshold is met, so one era is not judged under two different rules. Code and rollout conditions are verified; the required signaling and live activation were not independently confirmed.
SN94 Cathedral fixes an inconclusive AMD check being treated as a pass
SN94 Cathedral’s new probe code treats an unavailable AMD verifier as inconclusive, rather than a successful verification. The change addresses a documented rate-limit case where a tampered-signature control appeared to pass without the signature check running. It is a verification-tool fix, not an independent security certification or confirmation that unrestricted mining has reopened.
SN114 SOMA reports a Stage 2 sandbox failure and planned reruns
Today’s team-attributed brief says scaled sandboxes are failing across Stage 2 tasks, with runs ending in seconds and reporting zero token metrics. The team reports pausing the platform process and plans to fix the sandboxes and rerun affected evaluations. No completed restart or rerun result was located, so this remains an unresolved incident.
SN117 everyframe.studio releases downloadable everycli binaries
SN117 everyframe.studio has now published everycli v0.2.0 with downloadable macOS and Linux binaries and an installer, advancing beyond yesterday’s pending-release status. Recent versions add self-service onboarding, offer tables and account-credit checks for supported providers. The balance view shows remaining credits, not miner profit; optional snapshots are marked stale after 15 minutes. Publication verifies distribution, not that every miner is admitted or earning rewards.
SN120 Affine’s verification rewrite remains a nonpayable pilot
SN120 Affine publishes a synchronous pipeline that checks miner inference evidence before scoring and training the next checkpoint. Its records document remote trials, independent replay checks and subsequent trained-checkpoint sampling. October 1 work expands historical-family replay qualification. The current state explicitly separates these experiments from production and marks them nonpayable, with no chain-weight submission by the trial. This is inspectable engineering progress, not confirmation that the new pipeline has replaced the legacy reward system or improved model performance.
SN121 sundae_bar adds conversational directory search
SN121 sundae_bar demonstrates a directory users can search by describing what they want in a conversation. Filters update as the conversation proceeds, while results stay in the same grid. This is a specific interface feature beyond the broader agent overview already covered. Separately, today’s team-attributed brief reports an evaluation-gate bypass and a hotfix; no independent recovery or false-positive resolution was verified here.
SN122 CookingTAO plans SN50 Synth and SN22 Desearch mining integrations
SN122 CookingTAO identifies SN50 Synth and SN22 Desearch as its next mining integrations. The team says the new code is being screened before availability in beta 2, and promises further miner and revenue analytics next week. These are named upcoming integrations; completed deployment, new buybacks and audited revenue are not confirmed by the post.
SN124 Swarm previews Swarm Sentinel for drone-security simulation
SN124 Swarm says it is preparing to open-source Swarm Sentinel, an autonomous-drone security simulation environment aimed at sites such as solar plants and critical infrastructure. The October 1 post describes an upcoming release. No completed public release or deployed customer protection was established by this announcement.
SN8 Vanta reduces lock contention in status requests
SN8 Vanta merges a REST API change that reduces lock contention on read-only status operations and replaces subaccount structures consistently rather than modifying fields in place. This is a code-level response to status-request lag, with no independently measured production latency result.
SN49 Nepher livestream is scheduled for 1 PM ET today
Shizzy’s SN49 Nepher Robotics episode, “Can Bittensor Train Real Robots? Inside Nepher Robotics SN49,” is scheduled for October 1 at 1 PM ET / 10 AM PT. The YouTube watch page confirms the scheduled time. This is an upcoming conversation, not confirmation of a new model launch, tournament result or an already-started broadcast.
SN80 OpenRoboto joins Novelty Search at 5 PM ET
The OpenTensor Foundation schedules Novelty Search with SN80 OpenRoboto for Thursday, October 1 at 5 PM EDT / 21:00 UTC, hosted by Const. The announcement points to a discussion of the project; it is not evidence of another completed Shift product release.
SN98 NeverPlayAlone reports its dashboard is available
Today’s team-attributed brief says the SN98 NeverPlayAlone dashboard is now available. The source does not establish a new game launch or disclose verified revenue. No dashboard session was independently tested in this scan.
Audit trail
Coverage notes
Searched
- All 128 active subnet slots and 96 September 30–October 1 channel briefs reviewed.
- Signed-in X team searches and follow-up search for SN38 ChronoLLM, SN78 UMI and SN50 Synth read. 285 public account candidates sampled and 482 dated post records screened; candidates include quoted accounts and commentators, not just subnet teams.
- 176 repository paths scanned and 32 selected commit details inspected. New SN117 release assets, SN22 task API and SN78 competition API checked directly.
- Signed-in AlphaGap 119-subnet leaderboard, SubnetRadar, TAO Daily, TAO Media, primary partner statements and 52 dated Moments reviewed.
- September 24–30 editions checked for previous coverage, unresolved incidents and still-upcoming events. All-roster market outliers and leading emission recipients compared with selections.
- YouTube Videos and Live tabs checked; latest public upload title, date and 31:03 duration verified. Latest members upload and October 1 SN49 stream retained.
Public-data gaps
- Research cutoff 2026-10-01T15:10:00.467Z; includes material September 30 announcements after the previous edition's cutoff. X searches and public profile samples are not exhaustive; no Grok scan is claimed.
- 12 repository paths returned errors and 10 public account candidates exposed no usable posts. Some directory identities differ from current chain names.
- Channel briefs are AI-generated secondary summaries. Team-attributed operational claims stay developing; community reports and allegations do not prove deployment, cheating or recovery.
- No financial transaction, miner payout, contract, independent model benchmark or security certification was executed or verified in this scan. Code changes, published policies, testnet operation and production activation are distinguished.
- SN78 UMI rewards remain explicitly inactive in its public API. SN120 Affine's new verification trials are nonpayable; finalized current transition payouts were not reconciled. SN44 Score's Person-track recovery and SN114 SOMA reruns remain unconfirmed.
- SN55 NIOME's scoring investigation, unrestricted SN94 Cathedral mining, SN108 ChipForge activation, SN118 Ditto admission/reward recovery and SN88 Oops first conviction payout remain without new independent completion evidence.
- SN82 current identity remains unverified after re-registration. Its seven-day return spans the identity transition and is not treated as a catalyst for the former project.