Bittensor Daily News
Latest editionSN28 SayGM announces Gamma top-ups for AI model credits
SayGM announced on November 19 that users and agents can fund its AI service with Gamma. The stated workflow is to sign in with a Bittensor coldkey, mint Gamma on Subnet 28 and receive $1 of service credit per Gamma for Claude, GPT, Gemini and open-weight models. The team says existing API keys continue to work and no payment card is required. This gives Gamma a concrete customer-facing use: buying access to models through SayGM. The dollar amount describes service credit, not a guaranteed market price for TAO or subnet alpha. The announcement also says private routing applies across its models. The offering is team-announced availability. No wallet was connected or purchase executed during this review, and the linked Gamma instructions could not be retrieved. Actual customer spending and the payment flow have not been independently tested.
Around Bittensor
12 more storiesSN105 Beam offers Studio deployment inside a customer’s AWS account
Beam announced on November 18 that its Studio can be deployed into a customer’s own AWS account. Its documentation describes a CloudFormation-based setup using a dedicated EC2 instance inside the customer’s virtual private cloud, with encrypted storage and firewall controls. The change gives teams a deployment option in infrastructure they control, rather than requiring the workspace to run in someone else’s account. The documentation also covers backup and restore, workflow storage and agent access through MCP. This is Beam supporting AWS infrastructure, not evidence of an AWS strategic partnership or a disclosed customer contract. The announcement and deployment documentation were reviewed; a paid cloud instance was not launched.
SN8 Vanta reports its first trader-signal model trading live
Vanta’s team said on November 18 that its first model built from trader signals is live, and linked a public Hyperliquid wallet for observers to follow. The update describes scaling once the automation and model reliably run on a three-day cadence. This is a specific step from collecting trader signals toward deploying a model based on them. It should not be read as Vanta’s first-ever use of live capital: the announcement refers to this trader-signal model. The team’s statement and public wallet link establish the announced scope. Profitability, the model’s attribution to individual wallet trades and reliable long-term execution have not been independently verified.
SN47 Optia launches a platform for tracking brands in AI answers
Optia announced its launch on November 18, with Gordon Frayne identifying the project as Bittensor SN47. The public product offers a free scan of whether AI assistants name a business, which competitors appear instead and which sources the answers cite. Its stated workflow combines analysis, weekly monitoring and suggested changes to website content. The live product page lists a $49 monthly Monitor plan and a $99 Grow plan that adds more tracked questions and drafting tools. These are advertised product prices, not reported revenue. The launch and public offering are verified. A scan was not submitted, and the promise of improving a brand’s recommendations has not been independently measured. The example Allbirds dashboard is explicitly a public-scan illustration, not evidence that Allbirds is a customer.
Grayscale reports $1.05 million in Bittensor Trust share placements
Grayscale Bittensor Trust disclosed 231,800 newly issued shares in an 8-K signed November 18. Since the sales covered by its previous current report, the placements represented 4,366.89166140 TAO, valued at $1,045,408. The shares went to selected accredited investors at prices based on net asset value. The filing reports 3,479,100 shares outstanding as of November 18, with Grayscale Securities acting as authorized participant. Its earliest reported event is dated November 13. This is a private-placement update, not an ETF approval or evidence that the entire reported TAO amount was bought on an exchange on November 18.
SN91 Cascade reports a stronger GIFT-EVAL forecasting result
Cascade reported a 0.511 CRPS score on GIFT-EVAL on November 18, putting its mixture-of-experts system 0.005 ahead of Salesforce’s Moirai 2 on that reported measure. CRPS evaluates probabilistic forecasts; lower is better. The team separately reported 86.2% parity with Timestral on MASE, a different forecasting metric. Those measures should not be combined into a claim that Cascade beats every competing model across all tasks. The announcement described a competition that was 19 days old. This is a dated catch-up on the November 18 morning announcement, with a public stakeholder dashboard and forecast-access endpoint linked by the team. The results are team-reported and were not independently reproduced for this edition.
SN66 Conjectures reports a stronger bound for Erdős problem 18b
Conjectures announced on November 18 that it had obtained a stronger bound for Erdős problem 18b, publishing a paper and a linked solution. The team says the result is verified in Lean, a formal proof system. Its stated result says that, for every positive epsilon and sufficiently large n, every integer from 1 to n! can be expressed as the sum of fewer than n raised to epsilon distinct divisors of n!. This is the specific bound being reported, not a claim to have solved every related Erdős question. The announcement is a dated catch-up from November 18 at 10 AM Eastern. The paper and solution are available for scrutiny, but the formal verification has not been rerun independently for this report.
SN15 ORO finds three of 16 models beat a simple reorder rule
ORO highlighted its SMB-Bench results on November 18: three of 16 models in the expanded leaderboard beat a simple fixed reorder rule on average. The test simulates a wholesale buying desk across 90 mornings, with perishable inventory, suppliers, cash constraints and delayed consequences. These are simulated business outcomes, not profits earned by an actual wholesaler. The full report was published November 6, with the expanded leaderboard dated November 13. This is a catch-up on that research and its November 18 discussion, not a newly launched benchmark today. A scripted expert policy still outperformed the leading models, and average superiority did not mean a model beat the fixed rule in every run. The useful finding is the gap between noticing a business problem and consistently acting on it. ORO publishes results and traces for examination; the study has not been independently reproduced for this edition.
SN92 REAX 1.3 becomes available through Opper’s EU model gateway
Decision Models announced REAX 1.3 on Opper on November 18. Opper’s own provider page lists the model with a 29,000-token context window, EU hosting, zero data retention, no training on customer data and input pricing of $0.40 per million tokens. This is an additional distribution route for the existing REAX 1.3 model, not a new model release or evidence of paid customer adoption. The counterparty listing supports the availability and stated hosting terms; no inference request was submitted during the review.
Taostats adds decoy-wallet access and multilingual tutorials
Taostats announced on November 19 that its mobile app supports decoy wallets. According to the team, entering a separate PIN opens only a wallet chosen by the user. The announcement offers either a custom decoy PIN or the normal PIN entered backwards. The release follows a November 18 announcement that the website and app tutorials are available in multiple languages. Its initial wallet-setup tutorials cover English, French, Spanish, Chinese, German and Polish. These are team-announced product features, with official app-download links and demonstrations supplied in the threads. The app was not configured or independently security-tested for this review. A decoy display should not be read as a guarantee that other holdings cannot be discovered.
SN120 Affine retires Slots and pauses task rotation after its first round
Affine’s Slots page now states that the competition has been retired. This is a change from the live launch covered in yesterday’s edition; the public page does not give a reason for its retirement. Separately, the team reported on November 19 that round r000066 was its first completed round with task rotation: 32 reserve tasks were admitted and 2,000 tasks scored. Rotation was switched off at 13:09 UTC because the round’s public pool files were not yet reachable on affine.io. In a subsequent 10:05 AM Eastern update, the team said the public-file issue was fixed and scheduled a third attempt: round r000069, expected to open around 16:40 UTC. Normal rounds continue meanwhile; the restart remains a scheduled event, not a completed activation. The mechanism replaces tasks the leading model almost always solves with tasks from a public reserve. The latest update supersedes an earlier failed start that had swapped no tasks. The retirement notice and team’s operational explanation are verified; no unannounced cause for retiring Slots is inferred.
SN99 Axonry reports its first champion as mining opens
Axonry’s owner reported its first champion on November 19 at 9:42 AM Eastern, following the announced mining opening at Finney block 9,538,688, expected around 06:00 UTC. This is a new participation update after the project identity change covered yesterday. The published rules give the champion half of Axonry’s weight and burn the other half. Miners submit a proposed model-editing method, commit it on chain and then submit the matching files. The team describes a dedicated four-RTX-4090 sandbox for up to six hours, with GPU and agent usage charged to the entrant. A challenger must improve at least two benchmarks without meaningful regression on the others. The first-champion report and rules are attributed to the owner; this review has not independently reconstructed the winning evaluation or reward settlement.
SN116 Carbon continues grading fixes after queue bottlenecks
Carbon says it is still addressing grading bugs and improving speed on SN116. In its November 19 morning Discord update, the team said it wanted several smooth days of operation before judging the system’s health and capacity. The update follows a November 17 report of roughly 50 grades an hour and a resolved grading stall that evening. On November 18, the team said it had separated model building from grading, reused connections to reduce timeouts and added automatic retries for infrastructure failures. Those failures are not charged to miners, according to the team. Carbon also clarified on November 19 that a retry after an infrastructure failure keeps its original queue position and does not consume a new daily slot. A proposed threefold speed-up was part of the earlier upgrade plan; this review has not independently measured that gain or established full recovery.
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
Audit trail
Coverage notes
SayGM announces Gamma-funded AI credits. Beam supports customer AWS deployments, Vanta reports a live signal model and Optia launches AI visibility tools. Also: Grayscale share placements, three research updates, REAX on Opper, Axonry mining and Affine changes.
Coverage: Nov 18, 2026, 10:41 AM – Nov 19, 2026, 10:37 AM ET
Searched
- Bounded discovery across AlphaGap, SubnetRadar, TAO Daily, TAO.com, Taosis, 1t1b and Tao Outsider, with original-source follow-up.
- Official posts, team Discord messages, product documentation and research underlying the included stories. Grok discovery and attention were used only for leads.
Public-data gaps
- Full subnet, Discord, ecosystem-account, fund/person and media coverage remains incomplete. Additional leads and market-move causes are still being checked.
- eTAO and burns archive extension stopped at the runtime-review guard. Last-good archives are retained; a complete current event scan is not claimed.
- Owner-holdings refresh also stopped at its runtime-schema guard. Full Gamma transaction, lifecycle and service-use reconciliation remains incomplete.
Browse all 103 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.
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.
