Muse Spark 1.3·Gemini 3.8 Flash의 에이전트 경쟁, 실행형 개발 도구와 오픈 웨이트 확장

Meta has released Muse Spark 1.3, their fourth Muse Spark model release in five months. Muse Spark 1.3 (max), which is in limited preview for Meta’s partners, scores 62 on the Artificial Analysis Intelligence Index, behind only Claude Fable 5.1 and Claude Opus 5. The variant available now, Muse Spark 1.3 (xhigh), scores 61 and ties with GPT-5.6 Sol (max) and Grok 4.6 (high). Both variants’ gains come primarily from improvements in agentic work and scientific capabilities Muse Spark 1.3 (xhigh) enters the Artificial Analysis Intelligence Index at 61, up 4 points from Muse Spark 1.2 (57, August) and 8 points from Muse Spark 1.1 (53, July). It enters tied with GPT-5.6 Sol (max), Grok 4.6 (high), and Claude Opus 5 (high), and behind Claude Fable 5.1 (max, 66), Claude Opus 5 (max, 63), and Claude Fable 5 (max, 62) Muse Spark 1.3 (max), which is in a limited preview stage, lands at 62. This higher index score is enabled by gains vs. Muse Spark 1.3 (xhigh) in Tau3-Bench Banking (52% vs. 47%) and GDPval-AA v2 (1,754 Elo vs. 1,709). Muse Spark 1.3 (max) is second only to Claude’s Fable and Opus variants in total score Congratulations to @AIatMeta , @finkd , and @alexandr_wang on the release! Key Takeaways: ➤ Continued improvement on agentic knowledge work tasks. At the launch of Muse Spark 1.2, we noted its significant gains in agentic knowledge work performance vs. Muse Spark 1.1. The latest iteration continues this trend, with Muse Spark 1.3 (xhigh) demonstrating a notable 12-point gain vs. Muse Spark 1.2 in Tau3-Bench Banking (35% to 47%), a 5-point gain in Terminal-Bench 2.1 (80% to 85%), and a new GDPval-AA v2 Elo of 1709 against its predecessor’s 1615. Muse Spark 1.3 (max) improves further on Tau3-Bench Banking (52%) and GDPval-AA v2 (1,754 Elo). This Tau3-Bench Banking score is #1 among all models. Muse Spark 1.3 (max) achieves these higher agentic work scores by using more turns and total reasoning tokens, reasoning 62% more on GDPval-AA v2 and 28% more on Tau3-Bench Banking compared to Muse Spark 1.3 (xhigh) ➤ The lowest cost per task for any model at 59+ on the Artificial Analysis Intelligence Index. Muse Spark 1.3 (xhigh) costs $0.55 per Intelligence Index task at Meta's unchanged $1.25/$4.25 per 1M token pricing ($0.15 for cached input), with its peers GPT-5.6 Sol (max) and Grok 4.6 (high) costing $0.95 and $0.94 respectively, a 70%+ premium. This places Muse Spark 1.3 (xhigh) on the Pareto frontier for Intelligence vs. Cost per Task. Its cost per task is higher than Muse Spark 1.2 ($0.40 per task), driven by ~57% more input tokens per task on agentic evaluations, with output tokens up only ~8%. Pricing for Muse Spark 1.3 (max) is not yet publicly available ➤ Scientific Reasoning results rose across the board, led by CritPt. CritPt was the standout non-agentic score gain vs. Muse Spark 1.2, with a material +8 points for the xhigh variant (18% to 26%), and GPQA Diamond achieved +4 points (90% to 94%), while Humanity’s Last Exam and SciCode each gained a more modest 2-3 points (45% to 47% and 56% to 59%, respectively). Muse Spark 1.3 (max) achieved roughly similar scores to the xhigh variant, gaining 2 points in Humanity’s Last Exam, tying on GPQA Diamond, and losing a point on CritPt vs. Muse Spark 1.3 (xhigh) ➤ Minor regressions in only two evaluations. Both Muse Spark 1.3 (xhigh) and Muse Spark 1.3 (max) dropped 4 points in AA-LCR (83% to 79%) when compared to Muse Spark 1.2, and AA-Omniscience (Accuracy) fell 3 points for xhigh and 1 point for max. The drops in AA-Omniscience (Accuracy) are due to a higher abstention rate (not answering questions when unsure), which also lowered the hallucination rate for Muse Spark 1.3 (xhigh) Other model details (xhigh variant): ➤ Context window: 1M tokens, unchanged from Muse Spark 1.2 ➤ Pricing: unchanged from Muse Spark 1.2: $1.25/$4.25 per 1M input/output tokens, with cache hits discounted to $0.15 per 1M ➤ Input modalities: text, image, video ➤ Availability: Meta's first-party API and Muse Code

AI at Meta

@AIatMeta

8시간 전

We’re excited to release Muse Spark 1.3 with improved performance on agentic and coding tasks, and a focus on real-world usability. Key capabilities: → Sustains longer-horizon work across multiple workflows in a single thread → More actively collaborates with users: it asks clarifying questions, flags when it's stuck, confirms before consequential actions → Better calibrated on its own limits instead of hallucinating outcomes → ~20% fewer tool calls and ~25% fewer tokens vs. Muse Spark 1.2 in internal comparisons

Today we’re introducing a new 3.8 Flash model, our 3rd Flash release in just 6 wks. It delivers significant leaps from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning. On DeepSWE v1.1, it outperforms most larger frontier models in autonomously

Sundar Pichai

@sundarpichai

11시간 전

We’re also introducing Gemini 3.8 Flash Cyber, our most capable cybersecurity model. It shows frontier-level performance in discovering vulnerabilities and patching them at scale, with Flash-level speed & pricing. That includes achieving 86.2% on the important CyberGym industry benchmark, plus 47.2% on CWE-Bench for patching. We saw a 70%+ success rate in discovering vulnerabilities across 20 programming languages on our internal benchmark.

Karri Saarinen

@karrisaarinen

11시간 전

Our bug autofix loop in @linear has fixed 300+ bugs in the last 30 days. The team set it up as part of triage: it connects to Datadog and Sentry to investigate issues, then assigns them to Linear’s coding agent to fix

Cursor

@cursor_ai

6시간 전

You can now run Cursor cloud agents on your infrastructure, including pools of machines that automatically scale with demand. This lets you give agents access to internal services or specialized hardware, while the agent loop stays in Cursor.

I think it's unlikely that white-box techniques will be able to provide as much monitorability as CoT currently provides within a year. Currently, they are capable of catching some unverbalized thoughts / plans, but nowhere near as reliably as CoT seems to. I'm optimistic about "in a few years," since the rate of progress here is quite fast (our go-to activation decoding techniques for production monitoring were both published just in the last few months!). But even that is uncertain. Also, the mere existence of working techniques doesn't imply that they'd be widely applied, esp. if they are expensive or difficult to develop. I think we need defense in depth, and it'd be irresponsible not to work on developing + stress-testing white-box techniques, but it'd also be irresponsible to allow CoT monitorability to degrade substantially until we are confident we have a roughly-as-good working alternative. (To be clear, this isn't a dunk on OpenAI -- I have no evidence that OpenAI has allowed CoT monitorability to degrade substantially! Based on Jakub's comment, it sounds like probably not at this time). Aside: I think "mechanistic interpretability" isn't the right mental model for our current white-box monitoring approaches. It's really more like "mind reading" -- even when we can read the model's "thoughts," we're typically clueless about the underlying mechanism.

@nvidia is now the leading company in the number of open-source AI repositories on @huggingface . This chart tracks major Hugging Face repositories - models, datasets and spaces - over the past 12 months. A year ago, NVIDIA was one of several large contributors. It has since pulled away, adding more than 500 repositories, ahead of Alibaba Cloud, Hugging Face and Tencent since September 2025. These include its own Nemotron model family, Cosmos world models and GR00T data and models for robotics, as well as widely used robotics datasets and optimized versions of other models. This reflects a broader strategy recently articulated by @JensenHuang , who has increasingly positioned open models as central to NVIDIA’s role in AI. This massive acceleration is part of a broader shift in the AI ecosystem. Large technology companies are becoming central players in open-source AI. Leading figures such as @ylecun have long argued that open models could become a form of shared infrastructure, allowing companies, countries and researchers to build without depending entirely on a handful of proprietary models. What very few saw coming, even a year ago, was how quickly that view would become mainstream. The recent open letter signed by NVIDIA, Meta, Microsoft, Amazon, Google, and OpenAI and dozens of other leading AI organizations made the case explicitly - the AI ecosystem needs a strong open frontier. In 2026, open-source has moved to the very core of the AI ecosystem by pretty much every metric. Usage is perhaps the clearest signal. On OpenRouter, open models accounted for roughly a quarter of routed tokens through much of 2025. They crossed 50% in spring 2026 and have held a majority since May - meaning that, on the platform, more tokens now flow through open models than closed ones. At the same time, the performance gap with proprietary models has collapsed from years to just a few months. On most benchmarks, the best open-weight models score similar to the best closed-source ones. The leading open-source models are now Kimi K3, GLM 5.3, Qwen 3.8 Max and DeepSeek v4 Pro. Qwen and DeepSeek also lead in downloads on Hugging Face, while models from http:// Z.AI, Tencent, Xiaomi and NVIDIA are among the most used on OpenRouter. The center of gravity of open-source AI is clearly shifting, with Chinese labs - as we have noted for some time - now playing a defining role in both the technological frontier and global adoption. The more capable and accessible the open ecosystem becomes, the easier it is for companies to build agents, robots and specialized AI systems. That opens new possibilities for where value is created, who gets to participate in the AI economy, and how countries and companies think about technological sovereignty. *Note: We consider only repositories published by companies, with at least 500 downloads and 10 likes for spaces. Major non-company publishers, such as the nonprofit Allen Institute for AI (AI2), are therefore excluded.