Bittensor Toolkit →Find the right tool for your question.

Bittensor Daily News

Latest edition
#01 · SN85 · Vidaioverified
★★★★★

SN85 Vidaio launches updated video Studio with AI tools and integrations

Vidaio announced its updated product on November 17, including a browser Studio that identifies issues in uploaded footage, Pro upscaling, compression and object removal. The launch also offers an API and connections to AI assistants such as Claude and ChatGPT. The live product website describes a workflow for applying tools, comparing the result and downloading or sharing the output. Developers can integrate compression and upscaling into their own applications, while the assistant connector provides another route into the toolset. This is a product-release announcement, not just a redesigned homepage. The public site and original announcement were checked; a paid processing job, output-quality comparison and every advertised tool were not independently tested.

Why 4 stars: Four stars: a concrete expansion of the user-facing video product exceeds a cosmetic website refresh. Five would require stronger evidence of broad adoption or independently demonstrated performance.

#02 · SN99 · Axonryverified
★★★★★

SN99 identifies as Axonry and publishes a weight-editing competition

Axonry’s own website now identifies the project as Bittensor Subnet 99, replacing the unnamed slot in our previous report roster. Its public repository describes a competition in which AI agents improve language models by directly editing their weights. This verifies the current project identity, not a reconstructed registration or ownership-transfer event. Miners submit a method and commit it on chain before publishing it. Accepted entries run in isolated GPU environments, with edits compared against a current champion. The repository says a successful challenger must avoid meaningful regression and improve at least two benchmarks; the champion receives the subnet’s scoring weight. The website names Arbos as a partner and describes its GPU sandboxes. Those are project statements; this review has not reproduced the advertised research results or verified settlement of competition rewards.

Why 4 stars: Four stars: a verified subnet identity and concrete operating model merit the established major-name-change priority. Five is not supported by independently reproduced results or broad adoption.

#03 · SN23 · Trishoolverified
★★★★★

SN23 Trishool plans paid cyber inference and profit-funded alpha buybacks

Trishool has outlined Phase 4: a cyber-model business starting with paid inference for security teams, followed by miner-trained specialist models and eventually a larger model of its own. The team discussed the plan on November 17 and published its detailed Phase 4 article dated November 18. The first proposed service is an unrestricted version of GLM 5.3, which the team says should arrive in two to three weeks. Later miner challenges would aim to produce smaller, cheaper cyber models. Those capabilities and the launch timetable remain goals, not completed results. Trishool says 100% of its profits would fund alpha buybacks. It describes inference and model licensing as potential income sources, with Astroware continuing as a separate company that could license the models. The distinction matters: this is a proposed link between a subnet business and its token, not evidence that customer revenue or buybacks have already begun.

Why 4 stars: Four stars: a substantive change in product and commercial direction exceeds routine three-star roadmap maintenance. Five is not justified by a proposed launch and unexecuted economic model.

#04 · SN51 · lium.iodeveloping
★★★★★

SN51 Lium says usage revenue funded its recent buybacks and burns

Lium representative Fish said on November 18 that the team’s buybacks and burns totaled roughly $3.3 million over the preceding week. Asked explicitly whether the funding came from usage revenue or protocol emissions, Fish replied that it was all from usage revenue. That is a material distinction for SN51: the team is describing customer-funded token purchases, rather than treating emissions as customer revenue. The statement is directly attributable to the team, but the dollar valuation, revenue accounting and complete transaction set have not been independently reconciled. The separate circulating claim of an 82,159-alpha burn is still awaiting a verified event and execution record in this report. It is not included as a confirmed transaction in our burns archive. TAO used to purchase alpha must also be distinguished from TAO destroyed: buying and burning alpha does not establish that the purchase-input TAO was burned.

Why 4 stars: Four stars: a specific team statement connecting customer revenue to token purchases is economically material. Five is not justified without independent financial and transaction reconciliation.

#05 · SN53 · engyverified
★★★★★

SN53 Engy assigns 80% of miner rewards to baseline capacity

Engy announced a new reward model scheduled from epoch 23, November 18 at 08:00 UTC: 80% for baseline capacity and 20% for buyer traffic. Previously, the team says, all rewards followed traffic. The baseline pool pays enrolled servers for staying online and serving assigned models even when buyer demand is lower. Within that pool, eight-GPU B300 servers receive a 70% allocation and eight-GPU B200 servers 30%. Uptime is sampled every five minutes; an empty hardware class burns its share. The existing quality-gated traffic pool remains open to all miners. The published structure and scheduled start are verified; individual reward settlement has not been independently reconstructed. Separately, DeepSeek V4.1 Flash prices are due to double on November 21 at 00:00 UTC: input from $0.03 to $0.06 per million tokens, and output from $0.12 to $0.24. Cached-input pricing is unchanged. Engy attributes the adjustment to sustainable capacity and reliable service.

Why 4 stars: Four stars: moving most miner rewards from traffic to capacity materially changes operator economics, beyond routine three-star maintenance. Five is not supported by broader network consequences.

#06 · SN120 · Affineverified
★★★★★

SN120 Affine launches eight-slot model competition with 10% of emissions

Affine announced Slots live on November 17 at 8:04 PM Eastern. The new competition allocates 10% of SN120 emissions to its winning model, while the existing v2 mechanism retains 90%. Its public dashboard was operating during the November 18 review, showing all eight model slots occupied. Each slot uses one GPU and lasts eight hourly cycles. Registered SN120 hotkeys can join a queue; a replacement starts from a fresh copy of the current winner. At the end of its tenure, a challenger is compared with that winner on the same tasks. A challenger receives no reward until crowned. This offers a different participation route from simply changing the existing leaderboard. Affine also merged a separate evaluation-sampling change on November 17. The team reports that its offline comparison used roughly 12% to 15% fewer tasks. That is an evaluation-efficiency claim, not a corresponding improvement in model intelligence.

Why 3 stars: Three stars: a live, meaningfully different miner competition and evaluation update, above a minor two-star note. The standing routine-competition cap prevents four-star inflation.

#07 · SN124 · Swarmverified
★★★★★

SN124 Swarm merges Andorra Test Field task and validator safeguards

Swarm merged its Test Field update on November 17. The new drone task uses a map of its real test site in Santa Coloma, Andorra: take off, locate people, report their positions and land while respecting altitude and boundary limits. The implementation specifies 5% of rewards for Test Field and 95% for Sentinel. The same change hardens evaluation isolation, expands world-seed generation and adds automatic recovery for stalled validator connections. These are concrete code changes affecting task integrity and reliability, alongside the new competition. A merged implementation does not prove every validator has upgraded or that a successful physical field test has occurred. The team’s November 17 announcement described the first field test as still being prepared.

Why 4 stars: Four stars: substantive evaluation safeguards alongside a new task exceed routine three-star competition news. Five is not supported by a network-wide impact or demonstrated field deployment.

#08 · SN104 · TAOstatusverified
★★★★★

SN104 TAOstatus announces automatic agent selection for repository security audits

TAOstatus announced an automated repository-auditing workflow on November 16. Its official post describes submitting a Git repository, automatically assigning a security agent, assessing vulnerabilities and returning a report with severity levels and suggested fixes. The practical change is removing manual agent selection from the stated customer workflow. This is a dated catch-up on the product announcement, not a claim that it launched on November 18. The announcement establishes what the team offers; it does not establish that every finding is correct or that a customer scan provides a demonstrated exploit. No repository was submitted during this review.

Why 3 stars: Three stars: a concrete customer-workflow improvement exceeds a minor two-star note. Four would overstate the verified scope of automatic selection without a broader new capability established.

#09 · SN41 · Almanacverified
★★★★★

SN41 Almanac changes duplicate-agent scoring to one skill claim per cluster

Almanac merged an update on November 16 that limits duplication across forecasting agents. Its documented upload rule rejects an agent already held by another registered hotkey, using a normalized fingerprint rather than merely comparing file names. The scoring change groups sufficiently similar predictions and gives each cluster one skill claim, divided among its claimants. Allowance shares remain unchanged. The separate optional cluster penalty defaults to zero; it should not be described as an additional active penalty. This catch-up matters to miners because copying the same strategy across more hotkeys no longer produces the same skill-claim treatment under the new code. The repository also publishes a historical replay, but that simulation is not a measurement of current live payouts.

Why 3 stars: Three stars: a meaningful miner scoring and participation change exceeds two-star maintenance. Four would overstate the bounded demonstrated scope.

#10 · SN114 · SOMAverified
★★★★★

SN114 SOMA announces competition results and plans a higher burn allocation

SOMA announced its completed competition review on November 17, assigning the winning five hotkeys shares of 40%, 25%, 15%, 10% and 10% of that result’s weights. Its earlier announcement said the winning compressor would become the default after review; availability of that default was not independently confirmed. The owner separately proposed an allocation change in two weeks: burn from 11% to 19% and miners from 41% to 33%, with validators at 30% and owner at 18%. This is a future emission-allocation plan, not an already executed transaction burn or a reduction of TAO pool liquidity. For participants, the distinction is timing: the competition review is finished, while the proposed allocation change still lies ahead.

Why 3 stars: Three stars: completed competition results and a specified future allocation change merit useful coverage beyond a minor note. The competition cap applies, and the planned change is not execution.

#11 · SN97 · Ingotverified
★★★★★

SN97 Ingot expands terminal evaluation data and standardizes task formats

Ingot’s team said its evaluation update went live on November 17 at 2:01 PM Eastern, following a temporary pause. The announcement reports ten times as many terminal trajectories, each with more than ten turns, alongside format standardization and task-text fixes affecting software-engineering and science evaluations. Ingot rewards compressed copies of a reference model that preserve its behavior. A broader evaluation set changes how those submissions are compared; ten times more trajectories does not mean models became ten times better. The dataset-size increase is attributed to the team, while the public repository documents the evaluation system.

Why 3 stars: Three stars: a substantial evaluation-set expansion affects miner comparison and exceeds minor two-star housekeeping. Four is not warranted by a bounded evaluation change alone.

#12 · SN22 · Desearchverified
★★★★★

SN22 Desearch tightens request limits and reduces task-release overhead

Desearch merged a task-API fix on November 17. All request methods now count toward the per-IP read limit, including HEAD requests that were previously omitted. Excess polling is rejected before reaching Redis, and rate-limit counters move into memory. The change also avoids reading entire task payloads during release. The pull request reports a reduction from roughly 8–24 milliseconds to about 0.2 milliseconds of Redis time in a local test using production-sized tasks. That is a specific component test, not an end-to-end service benchmark. A team reply on November 18 said the API change was live; independent production load testing was not performed.

Why 3 stars: Three stars: meaningful request-control and task-release improvements exceed a minor two-star maintenance note. Four would overstate the scope of this bounded operational fix.

#13 · SN28 · SayGMverified
★★★★★

SN28 SayGM adds GPT-6.2 Astra and Sol to its model catalogue

SayGM announced support for GPT-6.2 Astra and Sol on November 17. Both appear in its current model catalogue, and the dedicated Astra page provides API instructions, provider rates and recent usage figures. Those measurements are reported by SayGM; this review did not perform a paid API call. The announcement quoted a 12.4% discount, while the current Astra page shows different rates across providers. Prices vary rather than representing a fixed saving on every request. SayGM also distinguishes its sealed gateway from the upstream model: frontier providers receive prompts under their own API terms. The listing does not mean OpenAI inference runs inside SayGM’s enclave. The announcement’s combined model link returned 404 during review, but the model catalogue and dedicated Astra page were accessible.

Why 3 stars: Three stars: a useful expansion of an existing model gateway. A routine catalogue addition does not establish a strategic partnership or a new inference architecture deserving four.

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

Vidaio launches its updated video Studio. SN99 identifies as Axonry, Trishool outlines cyber inference, Lium attributes buybacks to usage revenue, and Engy changes miner economics. Thirteen stories cover products, subnet identity and operational updates.

Coverage: Nov 17, 2026, 10:35 AM – Nov 18, 2026, 10:41 AM ET

Searched

  • Early discovery across AlphaGap, SubnetRadar, TAO Daily, TAO.com, Taosis, 1t1b, CoinAsia and Tao Outsider; bounded direct Discord and original-source follow-up.
  • Primary announcements, repositories and product pages for all included stories. Grok discovery and attention completed; omission reconciliation recorded separately.

Public-data gaps

  • Remaining full-roster Discord, ecosystem-account, fund/person, media and market-move investigations are incomplete; a roster entry does not mean every source was reviewed.
  • eTAO and burns archive extension remains blocked by the collector runtime-review guard. Last verified archives are preserved; no complete current event scan is claimed.
  • Owner-holdings refresh attempted but stopped at its runtime-schema guard. Last-good holdings remain dated; Gamma and lifecycle reconciliation remain incomplete.
Browse all 102 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