Bittensor Daily News
From the archiveSN92 REAX 1.3 records 74.0 on the independent Decision Index
On November 6, the public Decision Index 0.4.1 shows SN92 REAX 1.3 at 74.0, ahead numerically of GLiDE 27B at 73.4, while the full-results table marks both at tied rank =1. The Full score combines 60% private same-skill tests and 40% private new-domain tasks; public benchmarks are displayed but do not count. Open model files and a two-part engine are available. The model repository was last modified November 3, so today’s development is the independent result and announcement, not a newly uploaded model. The card’s internal 73.71 measurement is a separate public-suite result. No independent rerun, general intelligence superiority or commercial adoption is claimed. Following the morning announcement, shared node-derived snapshots show alpha/TAO rising from 0.019668351 at 15:00:12 UTC to 0.026876723 at 17:36:12 UTC, a 36.65% increase. That observed market response supports the story’s significance; timing alone does not establish that the news caused the move.
Around Bittensor
12 more storiesOpenTensor explains post-quantum wallets; the shared activation gate remains off
On November 6, the OpenTensor Foundation explained ML-DSA-65/Falcon signature options and optional key rotation. It says standard post-quantum accounts need no registration deposit, while rotating accounts permanently reserve 0.2 TAO; existing wallets remain supported. This explains code already covered in releases 487/488. It does not establish enabled mainnet signing: HashedAccountsEnabled was still false at finalized runtime 488 block 9,447,649, timestamp 14:31:24 UTC November 6. The reviewed implementation checks that shared gate for fixed and rotating post-quantum authorization. No wallet creation, fund movement or cryptographic audit was performed.
Shizzy publishes a new TAO episode; the Carbon livestream remains canceled
Shizzy published a new 19:04 public episode on X at 18:12:28 UTC November 6. It mixes his opening perspective with the TAO weekly chart, SN92 REAX's Decision Index result, the 10-to-100 TAO portfolio challenge and a rebalance from SN51 Lium into SN80 OpenRoboto through Mentat. These are his recorded views and portfolio discussion, not independently verified execution of every statement in the video. The post explicitly corrects the Carbon reference near the end: today's SN116 Carbon livestream is canceled and no replacement date is confirmed. The completed 37:26 SN71 Leadpoet interview remains the site's public default; the earlier 7:03 members update remains separately restricted. No full audiovisual review or completed Carbon replay is claimed.
SN36 Weir commits GPU price floor and model-linked activation fee checks
November 5 Eastern commits accept GPU prices between half the model cap and the full cap; under-floor GPUs are not rented and earn no idle pay. New activations remember the model, with a seven-day deadline to top up if a higher-fee model is reported. Insufficiently covered GPUs are withheld from earnings and rental; expiry resets activation, restoring prior payment as credit. Legacy activations without model records are exempt. Code and upstream tests were reviewed, without running those tests or verifying live rollout.
SN50 Synth reports an AI-built forecaster briefly led its hourly competition
At 17:25 UTC November 5, SN50 Synth reported that a model built by a Gemini 4 Argon agent topped its live Crypto1h competition the previous day, then fell outside the top 100 while its 24-hour ranking improved. Both the transient lead and reversal matter. This is a team-reported competition observation, not independently reproduced performance or a financial-return promise.
SN64 Chutes announces Qwen3.8 2.4T A95B at $2 input and $6 output
Late-found November 5 announcement, posted at 9:30 AM Eastern: SN64 Chutes says Qwen3.8 2.4T A95B is live, with text-only input, always-on thinking and a 262K-token context. The stated total is 2.4 trillion parameters with 95 billion active per token. Published pricing is $2 per million input tokens and $6 per million output tokens. The post distinguishes these open weights from the Max API’s larger context. This review verified the original announcement, not an inference/billing test; Max benchmark claims are not treated as measurements of this hosted endpoint.
SN66 Conjectures prepares a 307-target catalog after two withdrawals
November 5 PR194 adds 20 targets to the checked-in catalog, initially taking it from 289 to 309. Follow-up PR195 withdraws two targets after reported solved status, leaving 307 targets and 614 tasks; only open part (iii) of Erdős 12 remains rewarded. The verifier build, sandbox controls, dedicated worker and approved production rollout remain gates. This is a task-catalog change, not 20 solved problems or independent certification of the withdrawn proofs.
SN88 Investing88 announces MetaTrader onboarding live; Meta88 signals remain upcoming
At 16:30 UTC November 6 (11:30 AM Eastern), Investing88 announced MetaTrader live on SN88, following its prior-day launch plan. The team says Meta88, its separate trading-signals service, will launch after initial MetaTrader miner onboarding. The November 6 version 1.8.0 commit adds the new asset/PnL path; documentation says linked live or demo MetaTrader accounts participate through KYM, with daily scoring at 02:00 UTC. The public Meta88 preview is accessible, but KYM onboarding requires authentication and no linked-account workflow or signals delivery was independently tested. The attached diagram shows MetaTrader 5 → Investing88 rankings → Meta88 signals/copy trading; it is a product-flow graphic, not evidence of customers or returns. The November 5 weekly alpha-distribution statement remains a separate team claim; its transfers were not independently verified.
SN89 InfiniteQuant documents the end of HF and Markets, with Agent payments planned
The current master README says HF ended and Markets was discontinued effective November 5 at 00:00 UTC; new HF receipts are refused while earlier ones finish grading. It documents 30% of total emission for LF, 50% for Agent competition and 20% for referrers, with weekly agent-book payments planned from November 9 at 08:00 UTC. Those are documented competition policies, not a verified eTAO change or executed payment. Separate code adds 16 reduce-only daily changes beyond the normal 48 from November 7 at 00:00 UTC; deployed activation remains unverified. The effective date is explicit; the first announcement time was not established.
SN99 Thirty Spokes registration is replaced; successor identity remains unset
At 13:00:24 UTC November 6, finalized block 9,447,194, extrinsic 48 successfully removed SN99 and queued a replacement registration with a 252.775130692 TAO lock. Cleanup removed the Thirty Spokes identity at block 9,447,219; block 9,447,220 activated the replacement at 13:05:36 UTC. The successor has a different owner key and no identity set at today’s snapshot. This establishes registration replacement, not a negotiated sale or a successor product. Historical Thirty Spokes coverage remains labelled under its old registration.
SN118 Ditto announces a Home dashboard for conversations and agent activity
At 6:26 PM Eastern November 5, SN118 Ditto announced a new Home dashboard. Three inspected primary screenshots show conversation continuation, memory-theme cards, saved files and agent activity split into attention, completed and in-progress views. The screenshots use sample content. The team says the dashboard is live at app.heyditto.ai; authenticated end-to-end functionality and adoption were not tested.
SN120 Affine documents queue protection for turns left unjudged by infrastructure failures
Affine's November 6 documentation and public mining contract describe a no_decision outcome when infrastructure prevents a candidate from being judged, such as a hung checkpoint fetch or an expired probe. The stated policy leaves the hotkey queued through a K1 grant without credit or model promotion; a genuinely scored RED or GREEN still consumes its turn. The document records authorization at 12:36 CR (18:36 UTC), with a 2,400-second probe deadline measured from run start and a separate 900-second fetch bound. The sampled contract also distinguishes training's 32,768 output-token limit from the current promotion gate's 65,536, with context limits of 36,864 and 69,632 respectively. These are published operator rules and a dated contract snapshot, not an independent replay of a miner turn or proof that a subnet migration has completed.
SN121 sundae_bar opens SBC35 with a committed sealed reference set
The official Executive Presentation Strategy Skill challenge opened November 5 at noon Eastern and closes November 7 at noon Eastern. It tests a submitted skill across seven scenarios using a fixed base agent. Inputs, rubric and the SHA256 commitment to the sealed reference set are public during the challenge. Per-scenario/per-grader scores and integrity-gate notes are visible, but quality judges’ written explanations stay withheld until close, when the reference set and full validator outputs are to be published. This primary wording differs from a secondary digest suggesting all judge feedback is immediately available. Payout execution and commercial integration were not tested.
Quick briefs & watchlist
13 updatesSmaller developments and early plans. Verification labels distinguish confirmed information from developing claims.
SN8 Vanta: two previously archived burns enter today’s dated news
Two successful SN8 alpha burns already published in the permanent archive now join the full edition: 195.917929437 alpha at block 9,443,166 (23:34:48 UTC November 5), and 125 alpha at block 9,443,324 (00:06:24 UTC November 6, still November 5 Eastern). Each individually exceeds the 100-native-token threshold. They are dated carry-forwards, not new events in today’s extension and not counted twice. The complete new interval found no additional qualifying individual burn.
SN8 Vanta records a 222.945066151-alpha transaction burn
At 15:35:00 UTC November 6 (10:35 AM Eastern), finalized block 9,447,967, extrinsic 14, event 210 records a successful direct burn_alpha call destroying 222.945066151 SN8 alpha. The event and call amount agree, with the same Vanta registration before and after execution. The burner is different from the registered subnet owner, so this is not labelled an owner buyback. It is a new qualifying individual burn, separate from the previously archived events carried into this morning’s edition. The complete new scan through block 9,448,379 found no eTAO allocation events.
SN17 404—GEN merges opt-in concurrent native proof contexts
November 5 PR82 merges opt-in support for one to eight independent native prover contexts. They grow lazily as calls queue; the default remains one. Proof format, ABI and parameter pins stay unchanged, while each initialized context adds memory for its constraint system and proving key. The team reports local artifact tests; a published release and installation by miners were not verified.
SN28 SayGM releases gmcli 0.5.22 with registry-selected Phala KMS safeguards
The November 6 gm-miner changes make new confidential-worker creation select a non-expiring Phala key management (KMS) contract allowed by the registry, using its default or an explicit permitted override. Without an allowed explicit override, missing or invalid defaults stop creation, and Phala commands use an isolated working directory so a local phala.toml cannot alter deployment. At 18:19:13 UTC, the project published stable gmcli 0.5.22, moving these safeguards beyond the earlier development-only merge. The tagged source retains KMS selection and isolated Phala commands, and the release provides platform binaries. No binary was installed and no production worker deployment or exploit remediation was independently tested.
SN49 Nepher pins its simulation sandbox dependency to the installed prebundle
November 6 PR12 pins the simulation sandbox’s aiohttp dependency to the version bundled with Isaac Sim 6.1. It avoids replacing the shared copy to satisfy a newer Isaac Lab requirement and breaking extension symlinks. The small code diff was reviewed; a rebuilt image, live rollout or robotics task result was not verified.
SN51 Lium merges container-start and dependency safeguards
November 6 UTC merges tighten container-start timing, lock PDM dependencies, handle Redis failures without scoring them as zero and retry repository DNS through a backup helper network. Further fixes preserve customer-ordered disk capacity, expire stale throughput failures and distinguish platform filler shutdown from provider load. The changes and reported upstream tests were reviewed; no independent test run, fleet rollout or repaired customer outcome is claimed.
SN60 Bitsec announces BitIgnite referrals for Sentios Continuous Coverage
At 12:25 PM Eastern November 5, SN60 Bitsec announced BitIgnite referrals for Sentios Continuous Coverage. The original and its graphic describe 10% on an introduced team’s first and further purchases over six months. This is the team’s offer, not evidence of commissions paid or customer sales. No affiliate arrangement or endorsement by Shizzy is implied.
SN62 Ridges merges Docker storage metrics for validators
November 5 PR538 adds Docker backing-filesystem space and inode metrics, distinguishes them from root-disk measurements and supports a mounted-path override. Remote or inaccessible Docker storage no longer silently falls back to an unrelated local disk. The merge is verified; fleet deployment is not.
SN71 Leadpoet merges bounded admission-review recovery and prospective judge routes
November 5–6 Arena merges add bounded admission recovery, prospective judge routes, provenance-aware training exports and scored-round publication safeguards. PR1088 preserves saved scored rounds when publication reads are temporarily unavailable; genuine malformed evidence keeps its integrity checks. Earlier recovery and database migrations remain operationally gated, and a published/tested DeepSeek image is separate from canonical activation. These merges do not establish a recovered production round or new customer adoption. At 19:59:13 UTC November 6, PR1090 additionally preserves sanitized redirect destinations in consumed web-response evidence and validates callable Chat tools in internal training targets. This helps retain retrieval provenance and reject undeclared model tool calls; reviewed source is not proof of production rollout or independently executed tests.
SN107 Minos merges an optional local ComputeHorde SDK runner
PR41 merged at 00:54:31 UTC November 6, still November 5 Eastern. It adds opt-in CONTAINER_RUNNER=chsdk alongside the native default, translating validator Docker jobs and lifecycle handling through the ComputeHorde SDK. The implementation imports the SDK’s local Docker client; it does not establish remote distributed execution or a production rollout. Documentation and relevant implementation were reviewed, without installing or running the upstream software.
SN114 SOMA merges a fix preserving Claude Code’s native request format
Late-found November 5 PR244 pins off Anthropic-to-OpenAI translation for Claude Code benchmark requests. Its documentation explains that conversion drops cache-control fields, thinking signatures and system blocks; native Messages requests instead pass through to the upstream gateway. The code also sets medium reasoning effort. The merge is verified; a secondary account of reverted competition uploads and an extended deadline still lacks independently inspected original operational confirmation. No corrected benchmark score or deployment outcome is claimed.
SN123 MANTIS adjusts configured competition weights
November 6 UTC configuration changes set CADUSD, NZDUSD, CHFUSD and XAGUSD competition weights to zero and raise BURN_PCT from 0.45 to 0.52. This is a documented UID 0 allocation setting, not a verified transaction burn or eTAO event. The FLOW document weight moves from 1.178 to 1.020 while its stated 5% share stays unchanged. Live rollout is unverified.
SN127 Zils merges bounded GPU waits and overlapping model intake
November 5 merges bound GPU waits to 600 seconds, permit prefetching up to three claims with serial admission and a 60 GiB disk floor, and avoid consuming a daily attempt while another pool owns the GPU. Follow-up PR211 corrects the displayed d3-mini input cap to 8,192; it does not expand context. These are operator safeguards in merged code, with rollout and restored service unverified. On November 6, PR212 adds deterministic family-quota sampling when a test family has more labels than its allocated questions. Added regression assertions check prior draw fingerprints for existing cases; no production round or model-result change was independently verified. At 20:22:28 UTC November 6, merged PR214 adds each pool's latest finished intake to GET /soup, including graded/rejected hashes and deferred or refused claims identified by UID and committed hash. Only deferral reasons are published; other refusal reasons remain private. This lets miner-facing tools inspect the most recent intake between rounds. The implementation and added tests were read, but no deployed endpoint or live intake was independently exercised.
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
SN92 REAX reaches the top tied rank on the independent Decision Index; SN64 Chutes lists Qwen3.8 open weights. SN99’s Thirty Spokes registration is replaced, with its successor still unnamed. Plus wallet activation facts, SN121’s sealed-reference challenge, miner tooling and Shizzy’s Leadpoet replay and the cancellation of today’s Carbon livestream.
Coverage: Nov 5, 2026, 10:27 AM – Nov 6, 2026, 10:23 AM ET
Searched
- All 128 active slots reconciled against finalized owners, identities, market returns and sector evidence. Identity-linked default repositories queried over a seven-day overlap; individual access and pagination outcomes retained.
- All dated November 6 SubnetRadar briefs in the loaded index read as discovery. Signed-in Grok full-roster discovery and qualitative top-five review returned; selected originals, model cards, independent benchmark and challenge pages verified.
- Const Reborn, TensiaFDN and TensiaBot originals and reply views inspected; owner originals, reposts, replies and bookmarks reviewed as discovery where returned. Complete chain event scan, lifecycle review and 128 owner-wallet collection completed. Videos and Live tabs plus relevant watch pages checked.
Public-data gaps
- Repository access was limited by HTTP403/429 responses and some missing links; capped default-branch results are not exhaustive code coverage. Original X timelines were not exhaustively searched for every subnet.
- Original Discord evidence for several brief-only operating and safety claims remains unavailable; Affine migration/evaluation claims and other unresolved leads are retained for follow-up. No unverified misconduct allegation is published.
- Authenticated TaoStats /v1 cross-check lacks a configured API key. Public SN113 still displays Unknown. Independent benchmark reruns, hosted inference tests and full audiovisual listening were not performed.
- The final Grok omission request reached its usage limit after fresh discovery returned. Secondary leads were checked directly where possible; this does not establish exhaustive omission clearance.
Browse all 92 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.