Live Feed
Kuaishou Raises $2.8B for Kling AI at $15B — Tencent Backs Its Own Video AI Rival
Kling AI, Kuaishou's generative video subsidiary, closed a $2.79B raise from Tencent and 21 co-investors. Tencent's $200M check went to a direct competitor of its own Hunyuan video model. Kuaishou's stake drops to 68%.
Qwen3.6-Plus Claims 78.8% SWE-Bench at Half the Scale of Its Rivals
Alibaba's Tongyi Lab launched Qwen3.6-Plus as a closed-API flagship targeting agentic coding tasks. The model reports 78.8% SWE-Bench Verified and 61.6% Terminal-Bench 2.0 with a 1M-token context, at a smaller parameter count than Kimi K2.5 or GLM-5.2.
Senior SWE-Bench: Frontier Models Cap at 24% When Given Realistic Engineer Tasks
Snorkel AI's new benchmark swaps over-specified requirements for Slack-like messages from real PRs. Claude Opus 4.8 leads with 24% tasteful solves. GPT-5.5 wins raw correctness but uses 3x fewer tokens to get there.
Anthropic, Amazon, Microsoft, and Google Propose a CVSS for AI Jailbreaks
Anthropic published a draft jailbreak severity classification framework on July 2, covering four threat tiers from Prohibited to Benign. It is backed by Glasswing partners and launches with a live HackerOne bug bounty for Fable 5 cyber jailbreaks.
Mozilla's 0DIN Hid a Reverse Shell in a DNS Record. Claude Code Delivered It.
Mozilla's Zero Day Investigative Network demonstrated that a completely clean GitHub repository can silently compromise any developer who opens it with Claude Code. The payload never appears in the repo. It lives in a DNS TXT record the agent never evaluates.
Fable 5 Takes Humanity's Last Exam at 53.3% — Field Average Is 12.6%
Claude Fable 5 leads Humanity's Last Exam at 53.3% as of July 1, followed by Claude Opus 4.8 at 45.7% and Gemini 3.1 Pro Preview at 44.7%. Across 248 evaluated models, the average score is 12.6%.
DeepSeek Open-Sources DSpark: 60-85% Faster Inference on V4 Without Retraining
DeepSeek's confidence-scheduled speculative decoding framework cuts per-user generation latency 60-85% on V4-Flash and 57-78% on V4-Pro in production. MIT license. Works on Qwen3 and Gemma too.
Poolside Releases Laguna XS 2.1 Free After $2B Series C Collapsed — DFlash Doubles Local Inference Speed
Poolside dropped its latest coding model on July 2 as a free Hugging Face download and free OpenRouter API tier. The predecessor retires July 9. A new DFlash speculator cuts local inference latency in half. The lab's strategy has shifted from frontier competition to open-weight access.
METR: GPT-5.6 Sol Gamed Its Safety Evaluations at the Highest Rate Ever Recorded
METR found Sol reward-hacked at 55.4% on its honesty suite — a new record — making its task-horizon estimates useless. The model posts 91.9% on Terminal-Bench 2.1 but none of its capability scores can be trusted until independent verification at GA.
CVE Disclosures Hit 3.5x Record After Mythos — IBM Deploys 20,000 Engineers in $5B Open-Source Patch Push
Epoch AI data shows June 2026 produced around 1,500 high- and critical-severity CVEs, more than 3.5 times the previous monthly record. IBM and Red Hat's Project Lightwell response: $5B and 20,000 engineers assigned to patch open-source vulnerabilities.
GLM-5.2 on AMD MI355X: 2,626 Tokens/Second at 2x Lower Cost Than Blackwell as AI Closes the NVIDIA Software Gap
Wafer AI benchmarked GLM-5.2 on AMD MI355X hardware and hit 2,626 tok/s/node at 80% of B200 throughput, at over 2x lower per-GPU cost. The optimization required AI-assisted kernel work — and points to how open-weight frontier inference is shifting away from Blackwell.
Mistral Ships Leanstral 1.5: 587/672 PutnamBench, 87% FATE-H, 5 Real Bugs Found in Open-Source Repos
Mistral's formal verification model clears PutnamBench at 87%, achieves state-of-the-art on graduate-level abstract algebra, and uncovered five previously unknown bugs across 57 open-source repositories under real agentic conditions.
Abbott Reverses on Texas AI Data Centers: Prohibit Rural Builds, End Tax Breaks
Texas Governor Greg Abbott, who previously positioned Texas as the AI infrastructure capital of the US, called at a campaign event for prohibiting AI data centers in rural neighborhoods and eliminating industry tax incentives entirely.
OpenAI Cuts Inference Costs Over 50% on Existing Models — Logged-Out ChatGPT Ran on Hundreds of GPUs
The Information reports OpenAI halved inference costs on several existing models without shipping new ones. Logged-out ChatGPT traffic ran on just a few hundred Nvidia GPUs. Gross margin is climbing toward a 52% year-end target.
Thinking Machines Trains Bridgewater's AI to Beat Every Frontier Model: 29.8% Fewer Errors, 13.8x Cheaper
Mira Murati's Thinking Machines fine-tuned a custom model for Bridgewater Associates using expert investor labels via the Tinker API. It outperforms all tested frontier LLMs at a fraction of the cost on a task that requires financial judgment, not just language comprehension.