newsroom
$ANTHROPIC · China Just Built a Claude Mythos You Can Download (GLM 5.2)
now playing · $ANTHROPIC
$ANTHROPICbearish· highunknown
US government bans Fable 5 after NSA breach demo; $1T valuation at risk
Anthropic's most powerful model Fable 5 was banned by the US government after red-team exercises showed it breaching NSA systems in hours; this regulatory off-switch creates exist…
$ZHIPUbullish· highunknown
GLM 5.2 matches GPT-5.5 coding at 1/6 cost; open weights shift enterprise economics
Zhipu's GLM 5.2 achieves near-parity with GPT-5.5 and Claude Opus 4.8 on coding benchmarks (SWE-Bench, DeepSweet) at 5-7x lower API cost, and its MIT-licensed open weights enable…
$DEEPSEEKbullish· highunknown
Microsoft swaps Copilot LLM for DeepSeek; lab raises $9B at $50B valuation
DeepSeek has become the default alternative for enterprises: Microsoft replaced Copilot's LLM with DeepSeek, and the lab just closed a $9B round at a $50B valuation with the found…
$MSFT···bullish· mediumunknown
Microsoft replaces Copilot LLM with DeepSeek over OpenAI and Anthropic
Microsoft's decision to swap Copilot's underlying LLM from OpenAI/Anthropic to DeepSeek signals a major enterprise shift toward cheaper, high-performance open-weight models and re…
$NVDA···bullish· mediumunknown
Nvidia developing competitive open-source models alongside Blackwell/Vera Rubin hardware
Nvidia is building its own open-source model lineup that competes at the frontier, creating a vertically integrated moat where hardware advances (Blackwell, Vera Rubin) accelerate…
$AAPL···bullish· mediumunknown
Apple ships on-device Siri LLM running locally for free
Apple has deployed a functional on-device LLM for Siri that runs locally on consumer hardware at near-zero marginal cost, capturing the low-end use-case layer and pressuring cloud…
$KNOWLEDGE-ATLAS-TECHNOLOGY···neutral· mediumunknown
Zhipu AI parent trades at 1300x sales despite $107M revenue
Knowledge Atlas Technology (Zhipu AI's public vehicle in China) trades at $136B market cap on $107M 2025 revenue, a 1300x sales multiple reflecting extreme scarcity of Chinese AI…
$SAKANA-AIbullish· mediumunknown
Sakana AI's Fugu multi-agent orchestration beats frontier benchmarks via model routing
Sakana AI's Fugu system routes prompts across closed and open models (Claude, GPT, GLM, Kimi) using a judge model to select the best output, achieving higher quality at 30-50% low…
$OPENROUTERbullish· mediumunknown
OpenRouter Fusion API validates multi-model routing at 30-50% cost savings
OpenRouter's Fusion API implements the same multi-model orchestration architecture as Sakana, delivering 30-50% cost reduction versus frontier models by routing easy tasks to chea…