PulseAugur
实时 19:06:27
English(EN) Cognition releases SWE-2 coding model, achieving high accuracy scores and supporting long-running tasks # AI # AINews

Cognition 的 SWE-2 编码模型以高基准分数首次亮相

Cognition 发布了其新的编码模型 SWE-2,该模型采用专家混合(MoE)架构,拥有 2.8 万亿参数,每个 token 激活 1040 亿参数。据报道,该模型在 Terminal-Bench 2.1 基准测试中取得了 92.8 分,在 FrontierCode 等任务上表现强劲,但在处理更长周期的代理任务方面落后于 Claude Fable 5.1GPT-6 Astra 等竞争对手。SWE-2 已集成到 Cognition 的 Devin 编码助手产品中,但不能作为独立 API 或本地部署。 AI

影响Terminal-Bench 2.1 上设定了新的 SOTA,但突显了长周期代理任务中仍然存在的差距。

排序理由 Frontier-lab 模型发布,附带系统卡和基准数据。

在 Mastodon — mastodon.social 阅读 →

AI 生成摘要 · Google Gemini · 来自 7 个来源。 我们如何撰写摘要 →

Cognition 的 SWE-2 编码模型以高基准分数首次亮相

本文如何被排名

Signal score
0 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Frontier Release
Frontier-lab 模型发布,附带系统卡和基准数据。
Source corroboration
7 independent sources
Strong cross-source corroboration — multiple independent publishers covered this within the clustering window.
Topics
model release, product
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
11 days old
Aged out of breaking-news scoring windows; ranking reflects the durable signal from the full source set.
Coverage growth since scoring
+3 source(s) since last score
New sources have picked up this story since our last re-score. Score will update on the next scoring pass.

完整方法见我们的编辑标准

报道来源 [7]

  1. Hacker News — AI stories ≥50 points TIER_1 English(EN) · cdnsteve ·

    Cognition的SWE-2在Terminal-Bench 2.1上达到92.8

  2. Mastodon — fosstodon.org TIER_1 English(EN) · [email protected] ·

    Cognition的SWE-2在Terminal-Bench 2.1上达到92.8分 https://tokenstead.ai/models/swe-2 # ai

    Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1 https:// tokenstead.ai/models/swe-2 # ai

  3. Mastodon — mastodon.social TIER_1 Polski(PL) · aisight ·

    Cognition 新的 SWE-2 模型旨在以显著更低的成本匹配领先的编码系统,这得益于其独特的 RL 训练。该工具仍然保持其独特性

    Nowy model SWE-2 od Cognition ma dorównywać czołowym systemom kodującym przy znacząco niższych kosztach, dzięki unikalnemu treningowi RL. Narzędzie pozostaje jednak produktem zamkniętym, dostępnym wyłącznie dla użytkowników agenta Devina. # si # ai # sztucznainteligencja # wiadom…

  4. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    开发 Devin 编码代理的公司 Cognition 发布了其迄今为止最强大的编码模型 SWE-2。SWE-2 使用强化学习进行后训练

    Cognition, the company behind the Devin coding agent, has released SWE-2, its most capable coding model to date. SWE-2 is post-trained with reinforcement learning from Kimi K3, Moonshot AI’s 2.8T-parameter open model. The model scores 50.0% on FrontierCode 1.1 Main, within 1 poin…

  5. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    Cognition发布SWE-2编码模型,通过单次强化学习将开发成本降低高达64% #AI #AINews

    Cognition releases SWE-2 coding model, lowering development costs by up to 64% through single-run reinforcement learning # AI # AINews

  6. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    Cognition 发布 SWE-2 编码模型,实现高准确率并支持长时任务 # AI # AINews

    Cognition releases SWE-2 coding model, achieving high accuracy scores and supporting long-running tasks # AI # AINews

  7. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    Cognition 的 SWE-2 在 Terminal-Bench 2.1 上达到 92.8

    Cognition's SWE-2 achieves 92.8 on Terminal-Bench 2.1 Article URL: https:// tokenstead.ai/models/swe-2 Comments URL: https:// news.ycombinator.com/item?id=4 9646778 Points: 3 # Comments: 0 https:// tokenstead.ai/models/swe-2 # Tech # Technology # TechNews # AI # Gadgets # Softwar…