Industry’s first MLPerf™ Inference benchmarking across multi-vendor accelerators. Orchestrate workloads across mixed-vendor GPU pools without penalties. Cisco’s first heterogeneous GPU MLPerf™ Inference submission optimizes your data center unit economics.
75·ALong
n
news8h agonews
China Telecom AI Officially Releases Xing4.0-29B Agentic Large Model for Single-GPU Deployment
75·ALong
n
news8h agonews
China Telecom AI Officially Releases Xing4.0-29B Agentic Large Model for Single-GPU Deployment. Beijing, China, 21st September 2026, FinanceWire … Read More
The post China Telecom AI Officially Releases Xing4.0-29B Agentic Large Model for Single-GPU Deployment appeared first on FinanceWire - Financial Press Release Distribution, Finance PR.
75·ALong
n
news9h agonews
ATH: Why AI Training Clusters Face a Faster GPU Refresh Cycle
75·ALong
n
news9h agonews
Why AI Training Clusters Face a Faster GPU Refresh Cycle
75·ALong
n
news9h agonews
Why AI Training Clusters Face a Faster GPU Refresh Cycle
75·ALong
n
news9h agonews
Clichmont AI announces public sale of CLAI as European GPU infrastructure expands
75·AShort
n
news15h agonews
Introducing THIS / THAT Model 1.0. Trained by FLock.
Not another chatbot. A best-in-class open-source model built to decide.
94.1% success rate on our 68-question decision test, outperforming GPT-5.6, Deepseek-4.1-flash, Kimi K3, Jev, and more.
Just 1.9B parameters. Runs locally. Measured decision latency of ~31 ms on an RTX 5080 laptop GPU.
We’re putting decentralized AI to work: giving developers and AI agents decision-making capabilities they can run on their own hardware.
Your model. Your machine. Not every decision needs the cloud.
70·B+Long
m
meme16h agomeme
quote: Remember @pendle_fi' TAM is not crypto yield. It is every asset with a future cash-flow stream that needs price discovery, a hedge, or a fixed-rate buyer.
Private credit, tokenized equities/dividends, infra, money-market funds, GPU financing etc. Basically issuers can tokenise each separately, but do not want to build a yield market from scratch. Pendle wants to be the common liquidity/rates layer and this NGI+ move is the first credible proof that Pendle can become the forward rates market for private assets.
This is bullish for $PENDLE cuz there is an upside chain👇
(1) A credible issuer sees Pendle as distribution + price discovery
(2) MM's and fixed income buyers build liquidity around PT/YT
(3) PTs become lending collateral and balance sheet assets
(4) The next issuer lists cuz the market already has users and liquidity
(5) More volume and yield fees support the buyback/distribution engine
Pendle has already showed that permissioned instances and a curator model are part of the RWA scaling path and that is essential in my opinion cuz serious private market capital will not necessarily enter a fully permissionless pool.
If Pendle can serve both open and permissioned structures, it becomes much harder to displace.
h/t to @Blockworks for the data | Partners Group's Next Generation Infrastructure strategy is now live on Pendle as NGI+ (10 Dec 2026 maturity).
@PartnersGroup manages $186 billion across private equity, infrastructure, private credit, real estate and royalties for institutional and private investors worldwide.
NGI+ is the onchain token linked to the NAV performance of Partners Group Next Generation Infrastructure, the firm's open-ended fund that invests alongside that flagship program.
Private infrastructure of this calibre has historically been reserved for pensions, sovereign funds and family offices, and is now readily available through Pendle x @AssetoFinance, opening access to the fund's $1 billion AUM onchain.
In 2.5 years, the fund has returned 48.8% net with volatility below 2.5%, returns and yield that can now be priced and traded around the clock on Pendle. Fixed rate is also now available for anyone looking to a lock in a return, which at time of writing is at 19%, higher than the fund's historical performance.
This listing marks the point where private markets and onchain fixed yield meet at institutional scale. Pendle is proud to have brought a Partners Group strategy onchain. As the world's largest asset managers tokenise their funds, Pendle is the infrastructure that turns those assets into tradable yield, and we look forward to unlocking more of them alongside partners like Asseto 🤝
0·-Neutral
n
news9/20news
Bittensor Ecosystem Highlights :: September 13–20, 2026
This week’s biggest stories across Bittensor came from Conjectures, Score, Pareton, TAO(.)com, Good Morning and Almanac.
[ @conjectures_io - Subnet 66 ] Conjectures miners resolved six Erdős problems in five days, closing mathematical questions that had remained open for decades.
Erdős problems are a famous collection of open questions that mathematicians have been trying to solve for generations. All six proofs were formally verified in Lean, marking a remarkable run of mathematical breakthroughs for Subnet 66 in just five days.
[ @webuildscore - Subnet 44 ] Score just released Score Studio, their alternative to Roboflow for building computer vision systems.
Engineers can generate and label data, train and compare models, then deploy their applications from one workspace. Small teams can use Subnet 44’s public models or commission new ones from Bittensor miners through Vision Lab.
@manakoai also joined F/ai at Station F, a highly selective AI program that accepted just 26 teams. For Manako, it means direct access to top founders, advisors, and lab builders as the team pushes to turn its vision stack into revenue.
[ @Pareton_ai - Subnet 10 ] A Pareton miner’s optimization was merged directly into vLLM, one of the main open-source engines used to serve LLMs.
The change lets Qwen models process around 4% more tokens on the same GPU in some workloads. Work discovered through Subnet 10 is now improving AI infrastructure used far beyond Bittensor.
[ @taodotcom ] TAO.com launched Crosschain swaps, letting users buy or sell subnet alpha directly with USDC or ETH on Ethereum without manually bridging into Bittensor.
The protocol handles the full route in the background, with swaps settling in around 13 seconds on average. Multiple DEX aggregators have already started integrating it, opening dTAO markets to liquidity from outside Bittensor.
[ @say_gm_ - Subnet 28 ] Good Morning launched a public dashboard tracking how much revenue it uses to buy back and burn alpha.
Its first burn removed $43K worth of alpha from circulation, offsetting 13% of miner emissions for the cycle.
[ @almnc_ai - Subnet 41 ] Almanac is doing the same with its own public revenue dashboard.
They have bought back and burned $40,689 worth of alpha so far in September, bringing its cumulative total above $116K.
75·ALong
n
news9/20news
Bittensor Ecosystem Highlights :: September 13–20, 2026
This week’s biggest stories across Bittensor came from Conjectures, Score, Pareton, TAO(.)com, Good Morning and Almanac.
[ @conjectures_io - Subnet 66 ]
Conjectures miners resolved six Erdős problems in five days, closing mathematical questions that had remained open for decades.
Erdős problems are a famous collection of open questions that mathematicians have been trying to solve for generations.
All six proofs were formally verified in Lean, marking a remarkable run of mathematical breakthroughs for Subnet 66 in just five days.
[ @webuildscore - Subnet 44 ]
Score just released Score Studio, their alternative to Roboflow for building computer vision systems.
Engineers can generate and label data, train and compare models, then deploy their applications from one workspace.
Small teams can use Subnet 44’s public models or commission new ones from Bittensor miners through Vision Lab.
@manakoai also joined F/ai at Station F, a highly selective AI program that accepted just 26 teams.
For Manako, it means direct access to top founders, advisors, and lab builders as the team pushes to turn its vision stack into revenue.
[ @Pareton_ai - Subnet 10 ]
A Pareton miner’s optimization was merged directly into vLLM, one of the main open-source engines used to serve LLMs.
The change lets Qwen models process around 4% more tokens on the same GPU in some workloads.
Work discovered through Subnet 10 is now improving AI infrastructure used far beyond Bittensor.
[ @taodotcom ]
TAO.com launched Crosschain swaps, letting users buy or sell subnet alpha directly with USDC or ETH on Ethereum without manually bridging into Bittensor.
The protocol handles the full route in the background, with swaps settling in around 13 seconds on average.
Multiple DEX aggregators have already started integrating it, opening dTAO markets to liquidity from outside Bittensor.
[ @say_gm_ - Subnet 28 ]
Good Morning launched a public dashboard tracking how much revenue it uses to buy back and burn alpha.
Its first burn removed $43K worth of alpha from circulation, offsetting 13% of miner emissions for the cycle.
[ @almnc_ai - Subnet 41 ]
Almanac is doing the same with its own public revenue dashboard.
They have bought back and burned $40,689 worth of alpha so far in September, bringing its cumulative total above $116K.