Company / organization
Qwen
By and about Qwen
The Economist weighs Claude's J-space against 200 theories of consciousness
The Economist's cover briefing examines Anthropic's J-space in Claude Sonnet 4.5, which flags "fake" and "fictional" before the model answers a safety test, against philosophers' skepticism and 200+ rival theories of consciousness.
newsGPT-6 Astra scores 95 on EyeBench-V3, 37 points clear of the field
OpenAI's GPT-6 Astra scored 95/100 on adi's EyeBench-V3 visual-perception benchmark at max effort, 37 points ahead of second-place GPT-5.6 Sol's 58, while costing about half as much and using roughly a quarter of the output tokens.
newsEmad Mostaque pitches locally owned AI “Champions” founded at a $1 valuation
Emad Mostaque's Intelligent Internet unveiled "the Champion": locally majority-owned, public-benefit AI firms founded at a $1 nominal valuation, with a modeled $100m+$100m round reaching a $1bn valuation but no jurisdiction or funding committed yet.
news“You don’t run a model, you run kernels”: Ahmad Osman on the layer nobody benchmarks
Osmantic founder Ahmad Osman argues inference speed lives in the kernel layer, not the model — a Strix Halo test gives a same-model ROCm rebuild 5.7-6.8% more throughput — and published an eight-project path for learning it.
newsThe week's top AI papers say the harness, not the model, is the variable
DAIR.AI's ten-paper roundup shows scaffolding, not the model, drives the gains: Prime Intellect's Prime Agent lifts ARC-AGI-3 Best@1 from 30% to 95.5%, while a compaction bug quietly erases 90% of safety rules after five rounds.
newsAgent-generated kernels cut Qwen-Image serving latency 42.3%
Baseten engineer Brian Li reports an agentic kernel-development framework cut serving latency 42.3% on Qwen-Image and 15.2% on FLUX.2, atop an already human-tuned SGLang stack on NVIDIA B300 GPUs.
newsChinese labs converge on one architecture: 3:1 linear attention and a 2,048-token budget
Z.ai's GLM-5.3-Flash and Alibaba's Qwen3.8-Flash-Next, released a day apart, independently converged on the same recipe: 3:1 linear attention, a 2,048-token attention budget and four-branch gated residuals, while MiniMax dissents and keeps full attention.
newsThe Mismanaged Geniuses Hypothesis: a 4B model hits 100% where Opus 4.6 gets 76%
Alex Zhang, Zed Li and Omar Khattab's Mismanaged Geniuses Hypothesis argues frontier LMs are undermanaged, not undersized: a 4B model RL-trained on 32k-context tasks hits 100% on a 1M-context, 8-needle test versus Opus 4.6's ~76% and Gemini 3 Pro's ~26%.
newsMazeBench: the best coding agent collects 13 gems, four frontier models get zero
MazeBench's code-enabled leaderboard has GPT-5.6 Sol topping the board with 13 of 100 hidden gems, claude-opus-5 at 12% and claude-fable-5 at 11%, while grok-4.6, ox-alpha, glm-5.3 and qwen3.8-max all scored 0%.
newsReview: Ratel, context engineering for production agents
Ratel is an open-source context gateway that retrieves only needed tool schemas per turn instead of loading entire catalogs, using in-process BM25 search by default with no vector database required.
newsDarkbloom's idle-Mac inference network doubles to 499 nodes in 60 hours
Darkbloom, Eigen Labs' idle-Mac inference network, grew to 499 nodes (432 hardware-attested) in 60 hours, up from 389 a day earlier, with utilization at just 8%.
newsPoLar: Dynamically skipping or looping transformer layers boosts math accuracy by 60.9%
Researchers introduce Program-of-Layers (PoLar), a training-free framework that dynamically skips, keeps, or repeats transformer layer segments per input, achieving up to 87.8% on DART-Math with Qwen2.5-3B, without modifying base model weights.
newsTen papers, one theme: the agent harness moves into the training stack
DAIR.AI's weekly roundup of ten papers finds agent harnesses moving into the training stack: Microsoft's Agent Lightning lifts Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4% using just 6K training examples and a 3,500-line proxy.
newsMicrosoft proves an exponential retrieval gap, and ColBERT holds a 99% lead after fine-tuning
Microsoft Research India proved single-vector embeddings need exponentially more dimensions than multi-vector ones to rank correctly, and on their new ANDOR benchmark a small ColBERT model beats OpenAI's largest embedder 16x at Recall@2 zero-shot.
newsAlibaba's Scroll drops context compaction and beats the best long-horizon agent by 37.4 points
Alibaba researchers unveiled Scroll, a context manager that skips compaction entirely and has the model write Python to retrieve what it needs, scoring 94.8% on LongMemEval_S, 73.1% on BEAM_10M and 86.7% on LOCA_256K with Qwen3.8-Max.
newsMicrosoft's Thinkingbox grades agents on the database: 66.5% once, 47.5% every time
Microsoft released Thinkingbox and a 507-task Thinkingbox-bench that grades agents by checking backend database state, finding Claude Opus 5 tops single-attempt success at 66.50% but passes all 20 tries on only 47.53% of tasks.
newsAgents that post-train AI revise their strategy 2% of the time
A Tsinghua, Renmin and UESTC study of 1,338 AI agent transcripts post-training language models found agents changed strategy in just 2.1% of 3,557 consecutive run pairs, even as execution gains lifted benchmark scores from 10.41% to 23.0%.
newsDeepMind's Recirculation adds recurrence to a frozen Gemma3, cutting GSM8k error up to 20.9%
Google DeepMind's Recirculation feeds deep-layer activations back into a shallow layer one step later, cutting Gemma3's GSM8k error rate up to 20.9% on a frozen model with no retraining.
newsPhysics of Agents: an Ising model predicts LLM swarms with 75-86% accuracy
A Stanford team led by Batu El and James Zou found that a 100-year-old Ising model predicts how communities of LLM agents shift opinions in debate, hitting 75-86% balanced accuracy and beating every baseline.
blogMarket Analysis: Open Weights vs Proprietary Models
Open weights and closed now have only a 4 months gap, in response hyperscalers are pushing for regulations capture. Let’s examine how we got here and where this conflict is heading next.
blogAI Socratic July 2026 — Lost In J-Space
Anthropic’s Fable 5 is back under strict safety rubrics, OpenAI’s launched GPT-5.6, Meta launched Muse Spark 1.1 model and Meta Compute.
newsLiquidAI releases Antidoom, Open-Source Fix for Doom Loops in Reasoning Models
Liquid AI released Antidoom, an open-source method that eliminates "doom loops" in reasoning models by reducing failure rates from 10.2% to 1.4% on LFM2.5-2.6B and 22.9% to 1% on Qwen3.5-4B through targeted retraining of trigger tokens.
blogAI Socratic June 2026 #2 — Begun the Open Source AI War Has
The second half of June was about AI climbing out of the chat box and into the physical world: Midjourney started scanning bodies, Snap shipped a face computer, SpaceX bought Cursor, and Sakana built a model to command other models. Underneath it all, Dwarkesh Patel named the real bottleneck — the world refuses to be grindable.
blogAI Socratic June 2026 - Hoist by Its Own Fable
Anthropic shipped Claude Fable 5, its first public Mythos-class model, and 72 hours later a national-security directive pulled it offline worldwide. A company that spent the month lobbying to keep frontier AI pausable got its own pause, on schedule. Around it: new models from nearly everyone, a couple of S-1s, real math from the machines, and the usual carnival of vibe-coding pivots and rogue Waymos.
blogAI Socratic April 2026 — The Era of Mythos
Mythos, Claude Code leak, Anthropic surpass OpenAI on MRR
blogAI Socratic March 2026
Top AI updates from Jan 15 to Feb 15 2026
blogAI Socratic Dec 2025
The most important AI news and updates from last month: Nov 15 - Dec 15. GPT-5.2, Opus 4.5, Gemini 3, the Agentic IDE Wars, Genesis Mission, and more.
blogAI Socratic March 2025
All the most important AI news and updates from last month (Feb 20 - Mar 15).