Bittensor Toolkit →Find the right tool for your question.

Bittensor Daily News

Latest edition
#01 · SN24 · ms.computerverified
★★★★★

SN24 ms.computer announces hosting and AI-agent sandboxes

Ms. Computer announced its SN24 launch on November 15 at 2:20 PM Eastern, presenting HOST for free website or app hosting and RUN for an AI agent in its own sandbox. CONNECT, its inference-network offering, is explicitly marked coming soon. The announcement also invites users to a waitlist, so the launch should not be read as proof of unrestricted access to every service. The linked public miner repository describes software that runs customer workloads in an isolated sandbox on a miner’s host. Its documentation says validators check advertised resources and service delivery, with roughly four hours of qualification after enablement before a miner can earn. The software is proprietary and source-available, with permission to run it unmodified for SN24 mining. This moves the story beyond the earlier rename and owner-address coverage to an announced service offering. We verified the team’s announcement and published documentation; we did not deploy a customer workload or verify paid demand. In a November 16 Discord follow-up, owner nigirimaki said miners are currently paid according to the real work they perform, specifying that this work means running sandboxes. At 10:53 AM Eastern, he also reported that a brief intermittent mirror-login issue had been resolved. We did not independently test that resolution, and the discussion does not establish a cause for the token's price move.

Why 4 stars: Four stars: an announced hosting and agent-compute offering materially changes what SN24 provides, exceeding routine maintenance at three. Five is not established by the currently documented access and deployment scope.

#02 · SN80 · OpenRobotoverified
★★★★★

SN80 OpenRoboto announces machine.fun launch and Solana trading pair

OpenRoboto announced on November 16 at 10:35 AM Eastern that ROBOTO is coming to machine.fun with a trading pair on Solana. The team says this builds on its collaboration with peaq to give onchain identities to OpenRoboto robotics-data collectors. Peaq followed at 10:49 AM, saying ROBOTO goes live today paired with PEAQ. Peaq identifies OpenRoboto as Bittensor SN80 and describes Shift as paying people to record first-person footage of real work for robot-training data. Its follow-up says the Solana expansion is bridged through ForeverMoney using Chainlink CCIP. Those implementation details are attributed to peaq; we did not verify a bridge transaction or execute a trade. Peaq also says machines launching through machine.fun bond PEAQ for their working lives. Bonding locks tokens; it is not a burn. The announcement does not establish a new SN80 emission burn, trading liquidity, paid robotics demand or a change to Bittensor consensus. OpenRoboto and peaq both confirm the relationship and announced scope. The linked project page remained behind a terms-loading screen during our check, so independent trading-availability verification remains open.

Why 4 stars: Four stars: a counterparty-confirmed launch and cross-chain trading expansion materially broaden access to SN80 beyond a routine update at three. Five is not established by the announced integration scope alone.

#03 · SN90 · KubeTEEverified
★★★★★

SN90 KubeTEE reserves miner emissions for external operators

KubeTEE owner Pierre announced on November 15 that the subnet's Oakland-56 owner-operated cluster, UID 56, had been decommissioned. He said only external miner clusters would now be onboarded, scored and eligible for miner emissions. The owner's control plane, validator and platform services would instead be funded from owner emissions. In a November 16 follow-up, Pierre said external miners were already onboarded and SayGM remained connected and serving. He described changes to revenue recycling through Gamma as work in progress. The announced policy changes who can compete for miner rewards; it does not establish that the planned Gamma mechanism is live. The shutdown and onboarding status are attributed to the owner, rather than independently verified cluster or payout records.

Why 4 stars: Four stars: reserving miner reward eligibility for external operators is a substantive change to the subnet's operating model, beyond routine maintenance at three. The demonstrated scope is one subnet, without exceptional broader consequences establishing five.

#04 · SN92 · REAXverified
★★★★★

SN92 REAX team introduces a separate EU-hosted moderation beta

Decision Models, the team behind REAX, announced an EU-hosted Moderation API beta on November 16. It supports moderation rules written in plain language and offers a migration tool for existing Perspective API integrations. Its scores are independently calibrated, so developers need to retune thresholds and test their own examples before switching; the team says evaluation is pending and makes no accuracy claim. A November 15 update also introduced a setup page for its free, open-source local LLM autorouter. The tool classifies prompts and selects among configured models using pricing and benchmark settings. Users supply their provider keys and still pay the underlying inference charges; a proposed Pro version remains on a waitlist. These are developments in the team's product suite. Decision Models says EU-tier processing does not use the REAX miner network, so the moderation beta is not evidence of new SN92 miner usage or revenue. A November 16 image-classification example also demonstrates the existing s1-vision model rather than launching a new one.

Why 4 stars: Four stars: a new moderation capability with custom rules and migration tooling is more consequential than routine catalog maintenance at three. Exceptional adoption or ecosystem consequences supporting five are not established.

#05 · SN53 · engyverified
★★★★★

SN53 Engy reports 3.11 trillion inference tokens over 30 days

Engy’s public dashboard reported 3,105,679,412,402 prompt and completion tokens across the last 30 closed UTC days, October 17 through November 15. The dashboard snapshot was updated at 14:00 UTC on November 16. It describes the figures as buyer traffic across live catalog models, excluding Engy’s own probes. The period averaged about 103.52 billion tokens a day and peaked at 190.72 billion on November 13. DeepSeek V4.1 Flash accounted for about 83.5% of the total, showing that most throughput came from one model rather than an even spread across the nine listed models. These are operator-reported usage figures, not independently audited revenue or profit. Token throughput alone does not establish customer counts, margins or a fair token valuation. We have not carried forward the secondary article’s Venice valuation comparison because a comparable, independently checked measurement basis was not established in this review.

Why 4 stars: Four stars: a quantified multi-trillion-token usage milestone is substantial operational evidence for the subnet, beyond a routine model listing at three. Five is not supported by the reported scope alone or by an unverified valuation comparison.

#06 · SN117 · everyframe.studioverified
★★★★★

SN117 EveryFrame reports 46,733 jobs and a traffic-limit fix

In a report shared on November 16, EveryFrame said SN117 had run 46,733 AI media-generation jobs since launch, with 43,399 verified. Its figures were explicitly dated November 12 at 09:40 GMT. The report also listed 685 creator accounts across 56 countries and regions, 44 online miner nodes out of 81 registered, and 94.3% delivery reliability over the preceding 30 days. The team said the verified share of jobs fell to 81.9% on November 11 and 69.8% on November 12 after unusually high traffic reached a limit in its code. It reported identifying the problem and deploying a fix; the reviewed report does not establish post-fix reliability. These are operator-reported usage and account figures, not independently audited revenue or a count of paying customers. The article introduces a September 28 to November 1 reporting period but also includes later November activity, so its since-launch totals, rolling 30-day figures and daily results should not be treated as one identical measurement window.

Why 4 stars: Four stars: a substantive operating report quantifies completed work, account reach and a reliability issue with an attributed repair, beyond routine catalog maintenance at three. Its reported scope does not establish exceptional adoption or wider consequences supporting five.

#07 · SN120 · Affineverified
★★★★★

SN120 Affine reports AIME score rise from 60 to 90

Const said in Affine's Discord channel on November 15 that ten LoRA model updates and 30 million tokens took a nine-billion-parameter model from 60 to 90 on AIME in 33 hours, using public miner batches. The reported result describes an improvement produced through distributed training contributions. The post did not specify the AIME edition or full scoring and comparison protocol, and the result was not independently reproduced in this review. SN120's automated round notices continued on November 16, reporting another leading-model update applied at 14:43 UTC. Those notices track ongoing training rounds; their round-level pass rates are a different measurement from the AIME result.

Why 4 stars: Four stars: a specific, substantial reported model-improvement result from public miner training is more consequential than routine operating news at three. Incomplete benchmark details and no established exceptional wider result prevent a five-star conclusion.

#08 · SN3 · Teutonicverified
★★★★★

SN3 Teutonic lowers the threshold for challengers to replace the leading model

Teutonic's Tony announced on November 16 at 11:10 AM Eastern that the challenger threshold had been lowered. A challenger must now beat the current leading model with a lower confidence bound above 0.0005 bits per byte. The change adjusts the evidence of improvement required to replace the incumbent model. The quoted threshold is the team's announced rule, not a measured improvement in model quality or proof that a new challenger has already won.

Why 3 stars: Three stars: a changed evaluation threshold matters to participating miners and model selection, beyond minor housekeeping at two. The bounded rule adjustment does not by itself demonstrate a major model or ecosystem advance at four.

#09 · SN34 · Yugosverified
★★★★★

SN34 Yugos temporarily limits miner submissions to Bittensor providers

Yugos announced on November 15 that miner submissions would temporarily be restricted to Bittensor providers: SN64 Chutes, SN28 SayGM and SN53 Engy. Owner dubm said uploads would reopen shortly while v0.3.0 remained the active benchmark pack. He also said updates to the pack or its rules would reset the leading submissions. The team said it was limiting concurrent runs to avoid provider rate limits. Later that day, dubm reported work on a custom container queue because the provider's default queue could not reliably preserve first-in, first-out ordering. A completed queue fix or firm end date for the provider restriction was not established in the reviewed statements. The change defines which infrastructure miners can use; it does not by itself establish additional customer demand or a new benchmark result.

Why 3 stars: Three stars: changing eligible providers and submission handling materially affects miner participation, beyond small maintenance at two. It remains a bounded operating policy, without a demonstrated new product or adoption milestone warranting four.

#10 · SN116 · Carbonverified
★★★★★

SN116 Carbon expands Motor grading capacity after submission backlog

Carbon reported on November 15 that Motor 1.8 could grade two submissions at once and run compiled model builds. The team said the version and exam were unchanged, scores remained comparable and winning submissions would still be recognized in submission order. On November 16, it said more than 200 submissions had contributed to the grading queue and that repeat checks had been shortened. Later that morning, Carbon said it had purchased two additional grading machines and was moving model building onto A100 GPUs. Those additions were intended to clear the backlog; the team had not yet published measured throughput after the changes. Its earlier figures of roughly nine minutes for a typical grade and 28 minutes for the slowest tenth were explicitly measured before the upgrade, so they are not evidence of the new system's speed.

Why 3 stars: Three stars: grading capacity and queue handling materially affect participating miners, exceeding small housekeeping at two. This remains a bounded operating improvement, rather than a new product or wider network milestone warranting four.

#11 · SN75 · Hippiusverified
★★★★★

SN75 Hippius adds mobile uploads and sharing in Drive 1.3.0

Hippius announced Drive 1.3.0 on November 16, saying the update is live on iOS and Android. It adds uploads through other apps’ share buttons, drive sharing and invitations from a phone, and links from WhatsApp, Messages or email that open directly in the Hippius app. The change makes the existing storage product easier to use in everyday mobile workflows. This is an update to the app, not a newly demonstrated storage network or adoption milestone. The release details are attributed to the team; device installation and regional store availability were not independently tested.

Why 3 stars: Three stars: useful mobile upload and sharing capabilities improve an existing product. Two understates the workflow change; four would overstate this bounded app update.

#12 · SN97 · Ingotverified
★★★★★

SN97 INGOT reports fewer agent steps on bugs shared with Bonsai 2

In a November 12 result revisited in Dendrite's November 16 digest, INGOT said its latest leading model took 54 agent steps against Bonsai 2's 70 on SWE-bench Multilingual bugs that both models successfully fixed. The team described the comparison as 22% fewer steps at the same roughly 6 GB size. The comparison addresses how much work an agent does on that shared set of solved bugs, which can affect time and token use. It does not show a 22% improvement across every task, a higher overall solve rate or an independently reproduced benchmark. This is dated catch-up coverage of the November 12 team-reported result.

Why 3 stars: Three stars: a concrete, scoped efficiency comparison offers useful model-selection evidence beyond a minor update at two. Its limited shared-solved sample and unestablished broader results do not support a major general performance milestone at four.

#13 · SN114 · SOMAverified
★★★★★

SN114 SOMA details how it tests context compression and token savings

SOMA said on November 13 that its context-compression product could reduce an AI agent's token costs by 30%. A November 16 follow-up directed readers to its experiment methodology. This is dated catch-up coverage of the team's savings claim, rather than a newly verified result for every customer. The methodology compares paired runs using original and transformed context, with answer quality taking priority. It describes repeated tests and held-out examples, and a weighted token-cost calculation that counts ordinary input tokens at 1, cached input at 0.1 and output at 3. Those weights are a comparison model, not a universal provider price schedule. The methodology's illustrative charts do not independently establish the advertised 30% result. Real savings depend on the workload, model, caching and the amount of context that can be removed while preserving useful answers.

Why 3 stars: Three stars: a concrete efficiency claim accompanied by a defined evaluation method gives readers useful product evidence beyond a small update at two. The illustrative methodology does not establish a broad measured savings milestone sufficient for four.

#14 · SN66 · conjecturesverified
★★★★★

SN66 Conjectures expands its proof catalog and adds scoped Hilbert challenges

Conjectures expanded its mathematical challenge catalog through repository updates from November 13 to 16. The November 13 change added 19 Erdős targets and two reward-eligible Hilbert-related targets, taking the configured pool from 306 to 327. A Hilbert 16 target followed on November 15. The November 16 update adds three more narrowly defined statements related to Hilbert 6 and Hilbert 12, while withdrawing Erdős 345. The resulting configured reward pool is 330 targets, or 660 proof-and-refutation bundles. The public catalog lists 333 entries because it also includes entries outside that active reward pool. This gives miners more defined problems to work on; adding a target does not mean that the mathematical problem has been solved. The commits and catalog were checked, without an independent test of live reward distribution.

Why 3 stars: Three stars: the official repository expands the set of mathematical targets available to proof miners and updates reward eligibility. These are new challenges and configuration changes, without independently verified mathematical breakthroughs or a broader launch to support four.

Smaller developments and early plans. Verification labels distinguish confirmed information from developing claims.

#15 · SN99 · Subnet 99verified
★★★★★

SN99 Discord channel is claimed, with a project announcement planned for Wednesday

Bittensor moderator Kat confirmed on November 16 that Tobias had successfully claimed ownership of SN99's Discord channel. Tobias then said he would publish the project announcement on Wednesday, November 18. This confirms channel representation and a planned announcement. It does not establish an onchain ownership transfer, a sale, the project's operating scope or a relationship with another subnet.

Why 2 stars: Two stars: channel representation and a near-term announcement date are a useful bounded watchlist update. A substantive operating-team transition or product launch has not yet been established, so the higher identity-change or product ratings do not apply.

#16 · SN44 · Scoreverified
★★★★★

SN44 Score plans improved review of miner challenge annotations

Score's Mikhael said on November 16 that the team was putting an improved review process in place for challenges sent to miners. The statement followed a discussion of football action labels, including potentially missed pass and pass-receive annotations; it describes work in progress, without a confirmed rollout or measured accuracy improvement. In a separate November 16 post, Score described the operating constraints it uses before accuracy competition: a model must run in under 100 milliseconds on two CPUs and occupy less than 30 MB. These are the team's stated eligibility requirements, not evidence that the rules changed that day or an independently reproduced performance result.

Why 2 stars: Two stars: a planned annotation-review improvement is a useful bounded operating update. A completed change or measured effect on miner evaluation has not been established to support three.

#17 · SN56 · Gradientsverified
★★★★★

SN56 Gradients separates public subnet data from its training API

A November 14 update to Gradients' official G.O.D repository separates public read-only subnet information from customer training jobs. The code and documentation direct tournament, emissions and audit-data requests to api.sn56.io, while api.gradients.io remains the platform for creating and monitoring training jobs. The change also adds deployment instructions and API reference pages for the subnet service. This is a repository and documentation update; live endpoint health was not independently tested in this review.

Why 2 stars: Two stars: a concrete code and documentation change clarifies access to subnet data. It does not establish a first customer-API launch or a material measured operating improvement to support three.

#18 · Ecosystemverified
★★★★★

TaoStats adds German and Polish interface options

TaoStats announced on November 16 that German and Polish had joined its interface language options. Its stated six-language list is English, French, Spanish, Chinese, German and Polish, with selection through the globe icon at the top right of taostats.io. The team announcement establishes the added options; this review did not independently test each translated interface.

Why 2 stars: Two stars: additional interface languages make an existing Bittensor tool accessible to more readers. This is a product-localization update, without evidence of a new protocol capability or major adoption change to support three.

#19 · SN70 · Fugalverified
★★★★★

SN70 Fugal and Yuma schedule a November 18 smart-routing discussion

Yuma announced on November 13 that its Subnet Spotlight with Fugal will stream Wednesday, November 18 at 10 a.m. Eastern, or 15:00 UTC. The discussion concerns Fugal's approach to routing an AI query to the lowest-cost model capable of answering it correctly, with miners competing over those choices. This is an upcoming team discussion, not a newly verified performance result or product release.

Why 2 stars: Two stars: a confirmed upcoming team discussion gives readers a useful date and topic. The event notice does not establish a new routing release or measured cost savings to support three.

Stars indicate editorial priority, not investment quality. Routine competition updates receive at most three stars. Shizzy videos receive three; substantive non-video Shizzy coverage receives at least four. Scheduled events remain plans.

Sources & coverage notes
04

Audit trail

Coverage notes

Ms. Computer announces hosting and agent sandboxes, OpenRoboto expands toward Solana, and KubeTEE reserves miner rewards for external operators. Further coverage includes REAX's moderation beta, Engy and EveryFrame usage reports, Affine training results, Teutonic scoring, Yugos provider rules, Carbon grading, Hippius mobile sharing, INGOT and SOMA efficiency claims, and SN99's planned announcement.

Coverage: Nov 15, 2026, 12:39 PM – Nov 16, 2026, 12:08 PM ET

Searched

  • Direct Ms. Computer launch announcement, official product website and miner documentation.
  • Engy public usage dashboard, TAO Daily discovery article and Hippius original release announcement.
  • Bounded original-post reviews of OpenTensor, Bittensor and Const Reborn; signed-in Grok discovery, estimated-attention and omission queries.
  • Direct team-channel follow-ups in Bittensor Discord for SN3, SN24, SN44, SN90, SN99 and SN116; source dates and reviewed intervals are retained in the audit.
  • Decision Models dated changelog and moderation, model-routing and local-autorouter documentation.
  • Dendrite November 16 digest followed to INGOT and SOMA originals; late-found developments retain their November 12-13 dates. SOMA experiment methodology reviewed separately from its advertised savings claim.
  • EveryFrame original report and November 16 sharing post; statistics retain the November 12 cutoff and separate since-launch, rolling and daily measurement periods.
  • Affine original November 15 training-result statement and November 16 automated round notice; AIME and round-level pass rates are kept separate.
  • Yugos November 15 owner statements on provider restrictions, concurrent-run limits and ongoing queue work; active benchmark version distinguished from a new pack release.

Public-data gaps

  • Initial edition: full roster, outlet, seven-day social/media and fund-principal reviews remain incomplete; roster inclusion does not imply complete coverage.
  • eTAO and burns collection stopped at an unreviewed runtime version. Archives retain verified coverage through block 9513318; later events are not yet cleared.
  • Owner-holdings refresh stopped at the runtime-version guard. Last verified holdings are retained. Gamma and lifecycle event review remain incomplete.
  • Discord access was restored and the identified team statements were reviewed. This bounded follow-up does not establish complete channel, historical-watermark or market-trigger coverage. Remaining channel and price-cause investigations are recorded in the audit.
  • The repaired chart comparison passed all 384 checked timeframe results across 128 subnet slots. Registration continuity and the causes of flagged market moves remain separate unresolved checks; a passing chart comparison does not establish either.
Browse all 100 editions

Conference news · September 30, 2026

Exploit Conference Catch Up · Day 1

The teams, the ideas and the practical takeaways from Exploit Summit’s first day in Montreal. From Gamma and phone-based AI to robotics, science and the business of running a subnet.

Catch up on Day 1

Conference news · October 1, 2026

Exploit Conference Catch Up · Day 2

Catch up on Exploit Summit’s second day: Carbon’s Pitchtensor win, SN107 Minos’ research app, agent controls, robotics data, inference partnerships and the decisions behind building a subnet business.

Catch up on Day 2