Search “Reasoning”

DeepSeek open-sources inference framework DeepInfer

On October 20, 2026, DeepSeek open-sourced its internal inference framework DeepInfer. According to Yuchen Jin, citing data disclosed by DeepSeek, a single cluster processes 155 trillion tokens per day and generates $38 in revenue per GPU-hour (off-peak pricing); DeepInfer supports both NVIDIA Blackwell and Huawei Ascend. A subsequent 36Kr report stated that its accompanying open-source deepseek-recipe supports NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. On October 21, swyx, citing Latent.Space data, said a single cluster processes 155T tokens per day with a 96% input cache hit rate, and argued that the lifetime value of each sold GPU has risen by about 30%. On October 24, Nathan Lambert said DeepSeek's single cluster processes 150T tokens per day, exceeding the combined open-source model inference volume of OpenRouter, Fireworks, Together, and BaseTen, and noted that the actual number of clusters is unknown; on the same day, LMSYS published a same-hardware benchmark comparison, stating that SGLang meets or exceeds DeepInfer on DeepInfer's own benchmark, with prefill throughput up to 1.04x and decode step length shortened by 1.1%–7.4% on 64x GB300.

View story:DeepSeek open-sources inference framework DeepInfer

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.

English translation is being prepared.

Mistral releases Large 4 weights on Hugging Face

On October 29, 2026, Mistral released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter native multimodal model that continues reinforcement learning on the basis of the preview version, with improved coding and defensive cybersecurity capabilities. According to data published by Mistral AI, the model scores 19.7% on FrontierCode 2.0 and 48% on the Artificial Analysis Coding Agent Index, roughly on par with GLM-5.3, but uses an average of 47% fewer tokens per task; on Terminal-Bench 4.0 it improves by 13 percentage points to 41%. The model supports three reasoning modes, uses a modified MIT Mistral Ecosystem Open-Weight license, and is priced at $0.70 per million input tokens, $0.07 for cached input, and $2.10 for output via API. On the same day, Unsloth released a 1-bit GGUF quantized version of the model, stating that it can run locally on a Mac Studio, retains 91% top-1 accuracy after quantization, and provides a download link and usage guide.

Jensen Huang proposes that doubling model speed could double AI lab revenue

On October 29, 2026, Jensen Huang put forward a judgment: under the premise that demand exceeds capacity, doubling model speed (for example, through software or kernel optimization) could double AI lab revenue. The logic is that inference revenue equals the number of tokens served multiplied by the price per token; a faster kernel can increase tokens per second under the same GPUs, power, and data center conditions, and if all the additional tokens can be sold at the current price, a 2x speedup could double revenue with zero additional capital expenditure. Currently, this claim is presented in a report by Rohan Paul, with no other developments or official response yet.

The Institute for Advanced Study in Mathematics of Zhejiang University and Zhijiang Laboratory launch the Future Mathematics Program

On October 30, 2026, ScienceNet published an interview report mentioning that the Institute for Advanced Study in Mathematics of Zhejiang University and Zhijiang Laboratory jointly launched the Future Mathematics Program. The report stated that the program focuses on four areas: mathematical foundation models, formal verification, AI mathematical mechanisms, and interdisciplinary empowerment. Li Jianshu, an academician of the Chinese Academy of Sciences, said in the report that AI may change the paradigm of mathematical research, and pointed out that mathematics has done far from enough for AI; Wang Jian, an academician of the Chinese Academy of Engineering, believes that AI's reasoning ability currently comes mainly from code, and that learning deeper reasoning from mathematical data will bring greater gains. The report also mentioned that Google DeepMind's AI agent solved 9 previously unsolved Erdős problems and tackled 44 open mathematical conjectures in May this year, and that OpenAI recently published 722 mathematical manuscripts generated by its internal AI models on GitHub, while 25 Fields Medalists including Terence Tao and Deng Yu jointly issued an open letter warning of a serious misalignment between AI and mathematics. Currently, no further progress on the program has been disclosed.

Google releases world model Genie 4

On October 27, 2026, Google DeepMind released the world model Genie 4, focused on real-time interactive simulation and introducing native audio to the series for the first time. According to the official introduction, the model is built on capabilities from the Gemini family: Gemini Omni is responsible for making the simulation match the appearance and behavior of the real world, while other Gemini models provide the complex reasoning needed to maintain the simulation's coherence; users can control scenes in real time using natural language, and each world also instantly generates ambient sound, music, and interactive dialogue. The next day, Google DeepMind further explained that Gemini 3.8 Flash breaks down goals in a loop and checks frames to maintain causal coherence across multi-step events, and that Gemini 4 Argon has also been used to jointly simulate tasks with Gemini Robotics; the company said it remains an early experimental model, that object permanence and complex physical interactions still drift during long runs, that all generated content carries a SynthID watermark, and that it is currently being previewed to trusted testers first. On October 28, Ethan Mollick posted that Genie 4 can run DOOM, and tested it generating a game-like environment as it goes, saying the experience is not perfect. On October 29, Peter Yang used Genie 4 to build 5 worlds in real time, arguing that world models are the next big direction for AI after LLMs.

Nathan Lambert says funding for sovereign AI is underestimated

On October 28, 2026, Nathan Lambert commented that people underestimate how much funding can still be invested in the sovereign AI space. The comment cited Giotto AI's announcement of the release of two European sovereign AI models, Giotto Alto and Giotto Solo: Alto is aimed at advanced reasoning, is freely available, and is hosted on European cloud infrastructure, while Solo can run on a single GPU. There are currently no subsequent developments or facts.

OpenAI launches GPT-6.1 Terra on ChatGPT Work and Codex

On October 27, 2026, OpenAI released and launched GPT-6.1 Terra, touted for its speed and efficiency, visual reasoning, and precise computer operation, with clearer and more natural writing as well, rolling it out that day to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. On the same day, Arena announced that the model had entered Agent Arena and Code Arena; OpenRouter launched the model, saying it matches GPT-6 Astra on Agents' Last Exam at one-tenth the price per token; OpenClaw also announced that the model is now selectable. OpenAI co-founder Greg Brockman said Terra's computer use has crossed the threshold of a qualitative change and that using it is faster than operating a computer on his own. fal, meanwhile, used the fal Codex plugin to test turning Blender 3D content into Seedance 2.5 cinematic footage. On October 28, Arena announced that Terra ranked 14th in Code Arena: WebDev with a score of 1671, at a median cost of $0.40 per task, 28% lower than GPT-5.6 Terra. OpenAI's system card confirms that Terra is overall safer than older models, but its flag rates on items such as concealing uncertainty, circumventing restrictions, data exfiltration, and destructive operations

AMD Unveils CPU-Centric Helion Rack Design for Agentic AI

At OCP Global Summit, AMD announced Helion, a CPU-centric AI data center compute infrastructure design for running large numbers of AI agents simultaneously. Unlike Helios, which stacks large numbers of GPUs, this design targets the CPU demands created by agents repeatedly executing programs and accessing databases. A configuration for large data center operators can hold up to 32,768 sixth-generation EPYC Venice CPU cores in a single rack, with model inference requests handed off to separate GPU systems; the enterprise configuration places EPYC CPUs and Instinct MI350P GPUs in the same rack, holding up to 16 GPUs, and integrates a token router that routes requests to high-performance models or self-hosted open-source models to lower AI operating costs. AMD will offer Helion as an open reference design and contribute the specification to OCP; eight server vendors including Cisco, Dell, HPE, Lenovo and Supermicro are developing related systems, with launch planned for November.

Jassy says most AWS capacity is already booked through 2028

Amazon CEO Andy Jassy said that most of AWS's capacity is already booked through 2028. In an interview on the Andreessen Horowitz podcast, he revealed that AWS's large-scale compute capacity for 2026 has been fully booked, 2027 is essentially sold out, and a considerable portion of 2028 has also been locked in early by customers. Jassy mentioned that Amazon disclosed earlier this year that capital expenditure for 2026 will reach about $220 billion, mainly directed toward data centers, servers, and networking equipment; but he also said that data centers, power, and memory resources are tight, and the shortage may not end in 2027. He also said that 99% of AWS's top 1,000 EC2 customers already use its self-developed Graviton processor, that Trainium chips offer about 30% to 40% better price-performance than comparable processors, and that AI inference costs still need to fall by about tenfold to drive enterprise-scale adoption.

Nathan Lambert says DeepSeek processes 150T tokens per day on a single cluster

On October 20, 2026, DeepSeek open-sourced its internal inference framework DeepInfer. According to Yuchen Jin, citing data disclosed by DeepSeek, a single cluster processes 155 trillion tokens per day and generates $38 in revenue per GPU-hour (off-peak pricing); DeepInfer supports both NVIDIA Blackwell and Huawei Ascend. A subsequent 36Kr report stated that its accompanying open-source deepseek-recipe supports NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. On October 21, swyx, citing Latent.Space data, said a single cluster processes 155T tokens per day with a 96% input cache hit rate, and argued that the lifetime value of each sold GPU has risen by about 30%. On October 24, Nathan Lambert said DeepSeek's single cluster processes 150T tokens per day, exceeding the combined open-source model inference volume of OpenRouter, Fireworks, Together, and BaseTen, and noted that the actual number of clusters is unknown; on the same day, LMSYS published a same-hardware benchmark comparison, stating that SGLang meets or exceeds DeepInfer on DeepInfer's own benchmark, with prefill throughput up to 1.04x and decode step length shortened by 1.1%–7.4% on 64x GB300.

View story:DeepSeek open-sources inference framework DeepInfer

Evan Conrad says NVIDIA cannot build its own compute trading market

On October 21, 2026, SF Compute CEO Evan Conrad argued on the Latent Space podcast that NVIDIA cannot build its own compute trading market. His reasoning is that companies with large hardware or inference businesses find it difficult to operate a neutral exchange. In the same appearance, Conrad also said that the compute market must first achieve physical delivery—that is, actually delivering usable GPU compute—before discussing financialization; otherwise, market manipulation and basis risk are likely to arise. He explained that SF Compute is expanding from subleasing GPU clusters to building its own cloud operations capability in order to achieve physical delivery of GPU contracts, so that buyers receive usable compute rather than cash settled against an index; the company does not hold hardware in order to remain neutral, and it already supports customers in partitioning InfiniBand fabric online themselves. He also said that SF Compute has placed 5% of its capacity on the order book, that customers who spend $100 to buy capacity can on average recover $25 by reselling it, and that hundreds of millions of dollars in volume has already been traded. There are currently no subsequent developments.

Leash Bio releases Minos preprint, blog, and open dataset

Leash Bio has released the Minos preprint, blog, and open dataset. In an interview published by Latent Space on October 15, 2026, co-founder Ian Quigley described the company's work on extracting clean, reproducible protein–small molecule interaction data from DNA-encoded libraries (DEL), and mentioned that the team is turning years of accumulated data into tens of thousands of lab-in-the-loop simulations for agents to perform verifiable scientific reasoning; this work later developed into Minos and led to the release of the preprint, blog, and open dataset. The interview also mentioned that in its Kaggle competition, none of the roughly 2,000 teams exceeded random performance on unseen chemical libraries, and that Hermes, using only its own data, matched Boltz on public benchmarks. The latest state that can currently be confirmed is the aforementioned release.

Recursive releases open-weight local coding model Flint

On October 21, 2026, Recursive Superintelligence released Flint, an open-weight local coding model. According to related reports, Flint is a 40B mixture-of-experts model that activates approximately 2.5B parameters per token, is licensed under Apache 2.0, and its weights have been made available on Hugging Face; it was built from scratch in under 4 months by the automated research system Eureka Machine. On October 22, Hugging Face co-founder Thomas Wolf commented that Recursive researchers quantized the lm_head—which is usually recommended to be kept at high precision—to 4 bits, and the result differed from bf16 by only 0.7 percentage points. On the same day, OpenRouter launched Flint, saying it is aimed at coding agents, supports a 262K context, reasoning, and tool calling, and is free to use during the launch period; Unsloth released a Dynamic GGUF quantized version of Flint, saying it can run locally on devices with 24GB of memory. The model now offers multiple usage paths through open weights, hosting platforms, and quantized versions.

Musk says H100 rental prices are higher than factory prices, and payback period may shrink to 6 months next year

On October 21, 2026, in his latest interview, Musk offered a judgment: the rental price of Nvidia's H100 is now more expensive than when it was brand new from the factory, and he stated that it is not entirely impossible for the payback period of computing hardware to shorten to 6 months next year. He explained that SpaceX only signs short-term contracts when renting out computing power, because the useful intelligence produced per watt of electricity rises year by year, and short-term contracts are renewed at a higher price per watt upon expiration. He also mentioned that the team wanted to outsource part of its inference computing power, and after asking around about GB300, GB200, Google TPU, and Amazon Trainium, none were available; even H100s in overseas data centers had all been rented out. The related report also reminded that Musk is both a lessor of computing power and carries the valuation narrative of Starship and space computing infrastructure, and his original words also emphasized that a 6-month payback is not certain to happen. Currently, this judgment has no subsequent factual developments.

Meta launches Muse Wonder and Muse Wonder Full models

On October 13, 2026, Meta released the Muse Wonder series, comprising two models, Muse Wonder and Muse Wonder Full, which it called its most capable models to date, achieving SOTA on personal and professional agent tasks, covering professional domains such as law, finance, and health as well as frontier coding capabilities. Both models feature Active Memory and Active Focus, and were trained, evaluated, and released according to the Meta Superintelligence Scaling Framework. According to Artificial Analysis data, the two models scored 52 and 56 respectively on Intelligence Index v4.3 at maximum reasoning effort, higher than the 48 of Muse Spark 1.3 (max); Muse Wonder is already available on Meta's first-party API and Muse Code, while Muse Wonder Full is available only to Muse Code users. On the same day, Arena added the two models to Agent Arena testing, and Simon Willison released the llm-meta-ai 0.2 plugin supporting the new models. On October 14, Arena announced Muse W