October 10, 2026 · Published daily
Subnet News — October 10, 2026
SN100 Cortex dates HyperTrain for Monday, SN104 TAOstatus announces mainnet validation, and SN51 lium.io reports a bigger month of GPU rentals. SN48 Quantum Compute publishes its 2.0 overhaul, SN105 Beam opens room webhooks, and SN120 Affine makes training evidence more inspectable. Plus SN108 ChipForge validation fixes, miner-rule changes and operational incidents. Coverage runs through October 10 at 7:17 PM Eastern; later reads verify sources within that window.
- Coverage window
- Oct 9, 2026, 1:23 PM – Oct 10, 2026, 7:17 PM ET
- Subnets in coverage roster
- 128
- Subnets with updates
- 19
Biggest stories first
Today's ranked news
SN48 Quantum Compute publishes its 2.0 mining overhaul
SN48 Quantum Compute published its 2.0 release on October 10. The new miner documentation describes receiving quantum-circuit jobs, submitting them through a provider adapter, uploading results and answering subsequent status queries. This creates a path for provider-backed quantum execution rather than treating a repository release as proof that every miner owns quantum hardware. Crucially, the miner README still says “COMING SOON.” The implementation is published; open participation and paid mining activation are not established by that commit.
SN51 lium.io reports $1.4 million in monthly GPU rental spend
SN51 lium.io’s October 10 month-in-review reports GPU rental spend up 45%, active users up 16%, rentals up 106% and first-time renters up 32%. It also reports 15% more providers and 32% fewer rental failures. Opentensor highlights $1.4 million of GPU rental spend over 30 days. That is marketplace spend reported by the team and amplified by Opentensor, not profit, independently audited revenue or a guaranteed token-holder return. This is fresh operating data rather than a repeat of the previously covered buyback announcement.
SN81 Reliquary adds a general and tool-use training corpus
SN81 Reliquary’s October 10 environment release adds 113,355 prompts across 127,766 rows, with a pinned dataset and integration as a stage-A corpus source. The environment includes instruction-following, structured-output, clarification and function-call grading, while some conversational categories remain ungraded. The practical change is broader, inspectable training material and evaluation rules. This is published corpus infrastructure, not evidence that a newly trained model has already beaten external benchmarks.
SN100 Cortex schedules HyperTrain mining for October 12
SN100 Cortex’s October 10 thread dates the HyperTrain challenge for Monday, October 12, with 70% of emissions allocated to collaborative AI training and mining intended to open the same day. The first project is OpenDecision, described as a Jev alternative, with model training from scratch and architecture experiments. Web, desktop and mobile access to open models is a later ambition. This is a specific upcoming challenge and allocation announcement, not mining already open today or an independently verified on-chain reward change.
SN104 TAOstatus announces mainnet validation across two mechanisms
SN104 TAOstatus’s October 9 announcement says its validator is live on mainnet for LLM-key contribution and AI security audits. The first mechanism relays and evaluates contributed API keys; the second benchmarks autonomous agents before offering vulnerability reports for applications or repositories. The matching source change sets the default burn percentage from 95% to 100% during bootstrapping. An environment override remains possible, so a code default is not proof that every live emission was burned. The mainnet announcement advances yesterday’s unconfirmed launch target; independent full chain replay and customer adoption remain outside this check.
SN105 Beam adds room webhooks for agents and external applications
SN105 Beam announced room webhooks on October 10, giving external applications and agents a documented HTTP entry point into room workflows. Its documentation describes webhook members, requests and responses rather than requiring every integration to use the same agent framework. A separate demonstration shows Hermes and OpenClaw working with shared visual context. These are concrete integration surfaces beyond yesterday’s Hermes announcement; the demo does not independently establish enterprise customer adoption or universal interoperability.
SN108 ChipForge fixes winner resets while burn concerns remain open
SN108 ChipForge published two October 10 validator fixes: reset a local winner when the server’s baseline epoch changes, and use the server’s exact score-to-beat while limiting restored-winner payments to unpaid hours before its reward end. Separately, SubnetRadar’s dated Discord brief attributes a 100% emission burn and validation overhaul to a team member following an exploited winner. The code changes are directly verified; the incident account and live burn percentage were not independently reconstructed from chain data. There is no verified all-clear or firm restoration date. At the 23:17 UTC alpha/TAO snapshot, the subnet was up 22.17% over 24 hours and 160.73% over seven days; those moves do not establish a cause or resolution.
SN118 Ditto opens a business partner program
SN118 Ditto’s public partner page now offers consultants, agencies and service providers a share of a client’s setup fee and recurring monthly retainer. Partners can offer the service under their own brand; each client receives a scoped build, hosting and support. Shares, tiers and terms are individually agreed rather than a published universal commission rate. This turns yesterday’s forthcoming partner-program mention into an inspectable offer. It is a distribution channel, not evidence that new deals, profit or buybacks have already occurred.
SN120 Affine publishes finalized training evidence and selection accounting
SN120 Affine’s October 10 changes publish finalized training inputs, diagnostics, evaluation evidence and actual selection/exclusion accounting. Its signed history catalog links retained evidence from round 14 onward. The documentation separates this read-only evidence exporter from mining admission, reward rules and training itself, and distinguishes completed learning cycles from public mining cadence. The fresh run starts from Qwen2.5-Math-7B-Instruct; older runs remain historical records. This improves inspectability without proving a new state-of-the-art result or treating every accepted submission as a training example.
SN13 Data Universe fixes validation of fresh Reddit posts
SN13 Data Universe merged a validator-side fix after yesterday’s cutoff to stop rejecting fresh Reddit posts against an archive snapshot that does not yet contain them. That addresses a specific false-failure path in data validation. The merged release is verified; it does not show that every validator has adopted the fix or require treating a new miner release as the remedy.
SN44 Score: challenge pipeline outage reported
SubnetRadar’s October 10 Discord brief reports that SN44 Score’s data pipeline stopped creating challenges, including Worksite PPE tasks. The brief attributes an expected return at the next tempo to the team and notes one annotation block was removed. The underlying private conversation and subsequent recovery were not independently verified. This is an operational incident watch, not a confirmed current outage duration or an all-clear; unrelated miner allegations are not repeated as facts.
SN60 Bitsec names TWIF as a continuous-security client
SN60 Bitsec announced on October 9 that TWIF has signed up for Sentios Continuous Coverage of its codebase. The direct connection here is a named client for a Bittensor security service, not unrelated token-market coverage. The team’s announcement is verified, but contract value, actual service activity and independent confirmation from the named client were not established during this check. Coverage does not endorse the client or guarantee that its code is safe.
SN61 RedTeam: browser failures reportedly affect Ada-3 scoring
SN61 RedTeam’s October 10 dated brief describes Multilogin failures affecting Ada-3 submissions since October 8. A participant reports two of four runs missing; the brief attributes acknowledgment of a browser problem and a planned restart to a team member. It also says there was no rescore mechanism for affected submissions. These are secondary summaries, not an independently replayed scoring failure. Recovery and any compensation remain unconfirmed.
SN62 Ridges merges validator evaluation load balancing
SN62 Ridges merged validator scheduling and load-balancing work on October 10, with matching endpoints, queries and database changes. This concerns distributing evaluation work rather than a new model or another change to yesterday’s upload pricing. The implementation is in the repository; production adoption and measured queue-time improvements were not independently tested.
SN71 Leadpoet tightens arena publication evidence
SN71 Leadpoet’s October 10 arena changes bind incomplete-publication judge witnesses to the current round. This reduces ambiguity about which round’s evidence is used when a publication remains incomplete. The wider development window includes scoring and claim safeguards, but commit volume alone is not a business milestone. These are verified reliability changes rather than a new customer, investment or independently measured model improvement.
SN74 Gittensor expands rental visibility and fixes SparkInfer builds
SN74 Gittensor’s rental tooling adds per-box market status, pinned hotkeys and busy-state visibility. Separately, SparkInfer’s 0.6.33 documentation adds Swift-Qwen3.8-27B support; the following 0.6.34 fix restores linking when NVFP4 is disabled, including affected Windows binaries. These are distinct operational improvements to renting compute and running inference. Published benchmark comparisons depend on the tested model, hardware and workload and are not a blanket speed advantage.
SN75 Hippius: eight-day confidential-VM test reports 99.84% successful checks
TAO Daily’s October 10 report describes Marcus “Mogmachine” Graichen’s September 30–October 9 evaluation of four SN75 Hippius confidential-VM servers in France and the Netherlands. It reports 4,292 of 4,299 scheduled checks succeeding, with seven network-related failures and automatic recovery. That 99.84% figure describes the sampled checks, not the probability that a workload is secure or an independent guarantee against every attack. The article describes the service as beta. The original raw test report was not independently reproduced, so this remains attributed developing coverage rather than a newly verified universal security claim.
SN89 InfiniteQuant changes market windows and reward accounting
SN89 InfiniteQuant’s October 10 source changes add daily and weekly Bitcoin Up/Down windows anchored at 5 PM New York, carry retained losses forward per entity before paying above a high-water mark, and size shorter markets using the preceding three hours of sealed ticks. Another change arms the V4 oracle record for markets starting at or after 17:00 UTC. These are substantive competition and accounting rules in published code, not independent confirmation of deployed enforcement or a promise of trading performance.
SN118 Ditto announces a live campaign with SN93 Bitcast
SN118 Ditto’s October 9 post says its Stitch3 Creator Campaign with SN93 Bitcast is live, with creators producing posts about Ditto V2. That is a distribution campaign for a Bittensor product, separate from yesterday’s first-business-client announcement. The original Ditto statement was checked; independent campaign performance, spend and customer conversions were not established. The announcement does not turn views or creator posts into verified revenue.
SN127 Astrid: payout schedule and competition pause outlined in brief
SN127 Astrid’s October 10 Discord summary relays a message posted on behalf of the team: MS5 payouts are intended to finish October 15, with MS6 and MS7 starting by October 20 and running four weeks under Astrid Score. Competitions would pause after MS7 while the archives and scoring are reviewed, including a proposed buy-and-hold hurdle. The original delegated speaker and live payout execution were not independently verified. Treat these as reported plans, not completed payments or a finalized new scoring rule.
Audit trail
Coverage notes
Searched
- All 128 current SubnetRadar profiles, 99 dated briefs and 189 associated repository paths.
- 200 associated public X account samples and signed-in original thread checks as @ShizzyUnchained.
- Signed-in Grok broad discovery and focused omission review, explicitly including Const Reborn; direct Const profile, replies and dated search.
- Signed-in AlphaGap dashboard; SubnetRadar news and Moments; TAO Daily, TAO Media and primary documentation.
- Preceding seven editions, market outliers, emission leaders and unresolved incidents.
Public-data gaps
- Public account samples, signed-in searches and Grok do not prove exhaustive coverage of every SN number, alias or deleted post. Private Discord was not independently read; brief-only incidents remain developing.
- Published code does not prove adoption by live validators. Customer and benchmark claims are attributed; undisclosed commercial terms and actual burn configurations are not independently audited.