AI News Brief
1. GLM-5.2 Open-Weight Model Challenges Claude and GPT on Coding Benchmarks
Z.ai (formerly Zhipu AI) released GLM-5.2, an open-weight model that scores 62.1% on SWE-bench Pro, trailing Claude Opus 4.8 by just one point on FrontierSWE while narrowly beating OpenAI's GPT-5.5. At $1.40 input / $4.40 output per million tokens (roughly 1/6th the cost of GPT-5.5) with a 1M token context window, it tops the open-weight rankings and runs on domestic Huawei silicon. Semgrep benchmarks also show GLM-5.2 beating Claude Code on IDOR vulnerability detection (39% F1 vs 28-37%).
2. Together AI Raises $800M Series C at $8.3B Valuation as Open-Source Inference Crosses $1B
Open-source AI cloud provider Together AI closed an $800 million Series C on July 1, 2026, at an $8.3 billion valuation. Annual bookings crossed $1.15 billion, signaling that open-weight inference is now a billion-dollar business as enterprises shift away from closed proprietary APIs.
3. Meta AI Chief Says "Watermelon" Model Has Caught Up to GPT-5.5
Meta chief AI officer Alexandr Wang announced that Meta's next model, codenamed "Watermelon" (successor to "Avocado"), has caught up to OpenAI's flagship GPT-5.5 in benchmarks. A new version of Meta's Muse Spark model with advanced coding capabilities is expected "soon," with SWE-bench Pro results anticipated. Meta continues its open-weight tradition with the Llama family.
4. AI.cc Partners with Hugging Face to Bring 500+ Open-Source Models to Enterprise API
AI.cc announced a partnership with Hugging Face to offer 500+ open-source models through a unified managed API, eliminating the self-hosting barrier that has slowed enterprise open-source adoption. The integration provides managed API access, enterprise compliance, and unified billing across the Hugging Face model catalog.
5. Anthropic in Talks with Samsung for Custom 2nm AI Chip
Anthropic is in early talks with Samsung Electronics to manufacture a custom AI chip, joining OpenAI, Google, Amazon, Microsoft, and Meta in the race to control AI hardware. Anthropic has already hired chip engineers and has committed to $50B in US data-center buildout with Fluidstack. Nvidia still holds an estimated 74% of the AI chip market despite the custom silicon push.
6. White House Asks OpenAI to Limit GPT-5.6 Release; OpenAI Proposes 5% Government Stake
The White House has asked OpenAI to limit GPT-5.6 access to a small group of approved partners first. OpenAI previewed three GPT-5.6 variants (Sol, Terra, Luna), with Sol comparable to Claude 5 Mythos, but wider release is gated by government approval. Separately, Sam Altman is in talks to offer the US government a 5% stake in OpenAI - an unprecedented move in tech policy.
7. Meta's Cloud Push to Sell Excess AI Compute Sends Chip Stocks Tumbling
Meta is reportedly planning a cloud infrastructure business to sell excess AI compute, sparking a Wall Street sell-off in AI chip and infrastructure stocks. The move signals either overspending on AI infrastructure or a broader push into cloud services, and raises questions about whether hyperscaler in-house compute will reduce demand for Nvidia and other external chip suppliers.