Skip to content
K = 0.731Subscribe
Viewing history for 2026年10月20日 周二‹ Previous dayNext day ›AccelerationBack to now

Today’s AI news: editorial picks and emerging stories

2026.10.20 · Tue· The agent era

The universe is cooling.
Civilization is heating up.

One day, seen from outside time, and the stories that just moved civilization forward.

Subscribe to the daily briefingExplore the accelerationThe cost of intelligence
Get it as a 4K wallpaper →
The day's lead02:58MiniMax releases M3.1 Flash with stronger reasoning and lower task costsMiniMax released M3.1 Flash, a reasoning model scoring 42 on the Artificial Analysis Intelligence Index, up from M3’s 29, at less than half the cost per task. It supports text and images with up to one million tokens of context; API pricing is $0.15 per million input tokens and $0.60 per million output tokens. Artificial Analysis reports a 93 percent hallucination rate because it answers rather than declines uncertain questions; open weights are planned for October 23.Read the story →

Pressure gauge · Trending now

Full trending list
  1. 1DeepSeek Open-Sources DeepInfer; vLLM Reports Matching Its V4.1-Flash Decode Performance43.5
  2. 2MiniMax Releases M3.1 Flash With Higher Benchmark Scores and Lower Task Costs Than M310.4
  3. 3Anthropic Commits 5% of Its Shares to New Public Benefit Fund5.5

Latest picks

Models

New models and versions, open weights, and changes in capability and pricing, with their evaluations.

10/20Tue
  1. The Decoder76

    MiniMax releases M3.1 Flash with stronger reasoning and lower task costs

    MiniMax released M3.1 Flash, a reasoning model scoring 42 on the Artificial Analysis Intelligence Index, up from M3’s 29, at less than half the cost per task. It supports text and images with up to one million tokens of context; API pricing is $0.15 per million input tokens and $0.60 per million output tokens. Artificial Analysis reports a 93 percent hallucination rate because it answers rather than declines uncertain questions; open weights are planned for October 23.

    Why it matters

    The comparison shows how fewer output tokens can reduce task costs even when benchmark leadership is mixed, with MiniMax and Deepseek each leading on different measures.

10/18Sun
  1. 量子位76

    Zhipu releases GLM-Legal, leading open-weight models on two legal benchmarks

    Zhipu released GLM-Legal, an open-source legal model that reportedly leads open-weight models on Harvey LAB and LegalBench, with a 19.2% task pass rate and 86.23% accuracy, respectively. Built on GLM-5.3-Flash, it combines legal mid-training with reinforcement learning across tens of thousands of legal task environments. Official demonstrations cover contract review, evidence analysis, and cross-border privacy compliance; Zhipu is inviting legal professionals to apply for its Beta and plans to integrate related capabilities into AutoClaw.

    Why it matters

    Harvey LAB’s requirement to satisfy every scoring condition helps contextualize GLM-Legal’s reported lead, since its 19.2% pass rate also indicates that most tasks fell short of meeting all requirements.

  2. 量子位81

    DeepSeek releases DeepSeek-V4.1-Flash-1017 and opens Search and Fetch APIs

    DeepSeek released DeepSeek-V4.1-Flash-1017 across its website, app, and API, with reported improvements in coding and agent benchmarks alongside shorter task runtimes and outputs. According to the report, training specifically optimized the model for DeepSeek Harness, which requires version 0.2.1-rc.1 or later to fully use the improvements. DeepSeek also opened Search and Fetch APIs for finding webpages and retrieving their content as Markdown, priced at 5 yuan per 1,000 successful requests and temporarily free until November 1.

    Why it matters

    The reported reduction in task runtime and output length alongside improved benchmark results offers a concrete comparison with the previous version for developers assessing agent operating costs.

10/17Sat
  1. The Decoder80

    Deepseek-V4.1-Flash-1017 improves agentic coding scores and lowers reported task costs

    Deepseek released Deepseek-V4.1-Flash-1017 through its website, app, and API, with open-source weights on Hugging Face and ModelScope. In Deepseek’s own DeepSWE results, it scores 77.0 percent at just under $0.20 per task, versus its predecessor’s 74.2 percent at around $0.50; independent benchmarks are not yet available. Deepseek also opened its internally developed search engine to developers through Search & Fetch APIs, with each endpoint priced at $0.75 per 1,000 calls.

    Why it matters

    Deepseek’s reported task costs alongside benchmark scores provide a concrete comparison of coding efficiency, with the article clearly distinguishing company results from independent evaluation.

  2. DeepSeek69

    🚀 DeepSeek-V4.1-Flash-1017 is here! DeepSeek reports stronger benchmark performance, faster task completion, and shorter outputs than the previous version, describing a Pareto improvement in performance and cost. The model is now live on its website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. DeepSeek recommends upgrading to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model.

    Why it matters

    The comparison with the previous version connects benchmark gains to task completion time and output length, providing context for the release’s stated performance and cost improvement.

  3. The Verge · AI77

    A week with Nvidia RTX Spark laptops reveals performance strengths and practical compromises

    The Verge’s three reviewers spent a week testing five Nvidia RTX Spark laptops, finding capable graphics and generally workable Windows on Arm apps alongside high prices, weaker CPU performance than Apple rivals, and demanding local AI workloads. Sean Hollister’s Asus ProArt P16 lasted roughly 5 hours to 5 hours and 40 minutes during everyday work. Their local AI experiments encountered substantial battery drain, heat, slow responses, and uneven results, despite configurations offering up to 128GB of unified memory.

    Why it matters

    The hands-on comparisons show how large unified memory and capable graphics translate into everyday tradeoffs in local AI workloads, battery life, and software compatibility.

10/16Fri
  1. The Decoder78

    Google's Gemini 4 Argon is closer to the AI frontier than any Google model in months

    Google DeepMind’s Gemini 4 Argon scored 97.5 percent on ARC-AGI-3 using the Provider Adapter Harness at a cost of $4,800, behind GPT-6 Astra’s 99.9 percent. Its score fell to 49.8 percent at low reasoning in the Standard Harness, where API problems prevented completion of all games. Epoch AI placed Argon at 165 ECI points, close to Opus 5.5 and GPT-6 Astra, although early estimates ranked it eighth in math and sixth in software engineering.

    Why it matters

    The comparison shows how testing environments can change benchmark standings, providing context for interpreting Argon’s proximity to leading models alongside its weaker math and software engineering results.

  2. Aravind Srinivas70

    Aravind Srinivas says Gemini 4 Argon has set a new SoTA on the Wide-And-Deep-Research (WANDR) benchmark and is now available to all Perplexity users in fast search and Computer (agent) modes. https://t.co/0U671BhbMj The quoted @perplexity_ai announcement specifies availability for Pro and Max subscribers. https://t.co/n1TyI9RhBX

    Quoted postPerplexity@perplexity_ai

    Gemini 4 Argon sets a new record on WANDR. It is now available in Perplexity for Pro and Max subscribers. https://t.co/n1TyI9RhBX

    Why it matters

    The post identifies availability across fast search and Computer modes, while its broader access claim differs from the quoted announcement limiting availability to Pro and Max subscribers.

  3. Hugging Face Blog83

    Reflection releases Beam, a 501B open model for coding and agentic work

    Reflection released Beam on Hugging Face, a 501B MoE model with 23B active parameters per token, a 1M-token context window, and an Apache 2.0 license. The text-only model targets coding and agentic work, with BF16, FP8, and NVFP4 checkpoints and day-0 support in transformers, vLLM, and SGLang. Hugging Face provides deployment and post-training recipes alongside its own task evaluations and demos.

    Why it matters

    The deployment guidance distinguishes active computation from total weight memory, helping teams assess hardware requirements for running and adapting a sparse model with a long context window.

  4. NVIDIA AI77

    Reflection launched Beam, a 501B MoE model, with open weights under the Apache 2.0 license and 40+ ecosystem partners. NVIDIA AI says it was trained on NVIDIA GB300 NVL72 and delivers over 70K tokens/sec/GPU on Blackwell Ultra out of the box on day 0. Reflection’s quoted announcement includes BF16, FP8 and NVFP4 weights, a DFlash2 drafter, and Mirror CLI; the Beam 1.0 NVFP4 checkpoint is available at https://huggingface.co/reflection/Beam-1.0-501B-A23B-NVFP4.

    Quoted postReflection@reflection_ai

    Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: https://huggingface.co/reflection

    Why it matters

    The Apache 2.0 license, multiple weight formats, and released training harness give developers concrete information for assessing Beam’s suitability for their deployment workflows.

  5. 雷锋网82

    Kling 4.0 launches with 30-second native video generation and 10-bit HDR

    Kling AI launched Kling 4.0 on October 15, introducing native video generation up to 30 seconds, 10-bit HDR output, and claimed improvements in fast motion and complex camera movements. The model supports up to 15 combined reference assets, up to 10 keyframe inputs, and prompts up to 8000 tokens. Support for 4K, 1080p, and 21:9 output, repeated video extensions totaling up to 2 minutes, and API access are described as forthcoming.

    Why it matters

    The reference inputs, keyframe controls, and HDR output provide concrete context for assessing how Kling 4.0 could fit into professional video production workflows.

  6. Artificial Analysis79

    Meta has released Muse Spark Open, an 816B (42B active) open-weights model that Artificial Analysis reports scores 39 on its Intelligence Index, making it the #1 US open-weights model. Muse Spark Open (xhigh) scores 22 points above Muse Glimmer (high, 17), Meta's 30B dense open-weights model. https://t.co/jf31mQNINe

    Why it matters

    The comparison places Muse Spark Open 22 points above Muse Glimmer on the same intelligence index, providing a concrete reference for the reported capability gap between Meta's open-weights models.

  7. Hugging Face Blog74

    Granite Switch 4.1 Combines Chat, Decisions, and Speech in One Endpoint

    IBM released Granite-Switch-4.1-8B and -30B, combining chat, a general-purpose Decision aLoRA, a specialized Guardian aLoRA, and Granite-Speech-5.0-Turbo-CTC transcription through one hosted endpoint. IBM says the combined model uses approximately the same GPU VRAM as its base model, with adapters reusing the base model’s KV cache. Users can also compose custom Switch models from Granite Libraries adapters; an Adapter SDK for training adapters in Mellea is planned.

    Why it matters

    Shared KV caches and modular adapters offer a concrete approach to combining general and specialized decisions with transcription while keeping GPU memory use close to the base model.

  8. Hugging Face Blog87

    Meta releases Muse Spark Open, an open-weight multimodal mixture-of-experts model

    Meta’s Muse Spark Open is an open-weight mixture-of-experts model with approximately 816 billion parameters, around 42 billion active per token, and a context window of up to one million tokens. Trained for agentic coding, it accepts text, images, audio, and video and generates text. Hugging Face provides inference and deployment recipes, reports its own quantization tests, and demonstrates attention-layer LoRA fine-tuning with TRL while keeping the experts frozen.

    Why it matters

    The deployment recipes clarify how sparse expert activation reduces per-token computation while retaining substantial memory requirements, helping developers assess hardware needs across checkpoint formats and serving frameworks.

  9. The Decoder78

    Black Forest Labs releases FLUX 3 Image Klein with 8 GB VRAM support

    Black Forest Labs has released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model for text-to-image generation, image editing, and generation with up to ten reference images. It supports native 4K output, and its INT8 build runs on NVIDIA GPUs with 8 GB of VRAM, generating an image in under 9 seconds on an RTX 4070. The FLUX Kommunity License allows free commercial use for companies making under $5 million annually.

    Why it matters

    The hardware requirements and commercial license threshold provide concrete context for assessing whether local image generation fits a team's available GPUs and business size.

  10. Mustafa Suleyman72

    We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results coming soon—stay tuned! Lots of details in our technical blog post. https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/

    Why it matters

    The announcement places MAI’s reported progress at the pre-training stage, with post-training results still forthcoming, providing context for interpreting its claim of state-of-the-art base models.

  11. The Decoder81

    Nvidia releases Nemotron 3.5 Super VL for continuously running enterprise AI agents

    Nvidia released Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters and 12 billion active per token, for continuously running enterprise agents. Nvidia says it approaches Nemotron 3 Ultra’s accuracy on specified agent tasks with about one-fifth the parameters, and is releasing weights, datasets, and training recipes. The accompanying Nemotron 3.5 Bolt 74B-A9B 24GB targets local agents, while Bolt E4B 4GB is planned for later this month.

    Why it matters

    The deployment details clarify how quantization, caching, and hardware requirements shape the practical tradeoffs of running continuous enterprise agents on company infrastructure.

10/15Thu
  1. Mistral AI77

    Mistral AI releases Timestral, a family of time-series foundation models

    Mistral AI released Timestral in four sizes, from 5M to 209M parameters, for probabilistic, multivariate forecasting with past-only and known-future covariates. Mistral reports that its 209M model ranks first among pretrained, zero-shot models across five metrics on GIFT-Eval, TIME and fev-bench, using a single checkpoint without ensembling. All sizes support CPU inference, and Mistral published a GitHub repository for running Timestral on users’ data; domain adaptation and fine-tuning are available through Mistral Forge.

    Why it matters

    The comparisons across model sizes and CPU latency provide a concrete basis for assessing forecasting accuracy against compute requirements in production deployments.

  2. TechCrunch · AI77

    Nvidia launches open models for enterprise agents and local AI PCs

    Nvidia launched Nemotron 3.5 Super VL, a 120-billion-parameter open-weight model for enterprise agents, alongside lightweight Bolt models for local use on AI PCs and workstations. Nvidia says organizations can post-train Super VL on their own data and deploy it on their own infrastructure, while Bolt models support locally run agents. Super VL and Nemotron 3.5 Bolt 74B-A9B 24GB are available now; the smaller conversational model, Bolt E4B, is due later this month.

    Why it matters

    The enterprise and PC model lineup illustrates how deployment location shapes agent offerings, with customization on company infrastructure and local processing addressing different needs for control over data and workloads.

  3. NVIDIA AI76

    NVIDIA AI congratulates StepFun on Step 5 Preview, calling it StepFun’s largest open weights release. NVIDIA says the 600B MoE model supports agent tasks lasting up to 24 hours and runs at nearly 8K tokens/sec/GPU on NVIDIA GB300 using NVFP4. StepFun’s quoted announcement lists 27B active parameters per token, a 1M-token context window, native text and vision input, and BF16 and FP8 weights on Hugging Face and ModelScope. NVP4 checkpoint: https://t.co/lSECChAwlW Deploy with https://t.co/GO1wqTSBOk https://t.co/28TBzUo6ZU

    Quoted postStepFun@StepFun_ai

    Step 5 Preview is now open weights. BF16 and FP8 weights are live on Hugging Face and ModelScope, together with our technical blog. • Sparse MoE: 600B total, 27B active per token • 1M-token context window • Native text and vision input • Built for long-horizon agentic https://t.co/KNmllYjwdC

    Why it matters

    The reported throughput ties Step 5 Preview to a specific GPU and precision format, giving readers concrete deployment context for its long running agent workloads.