GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 827 -5.3%
QWEN-38X 824 —
CL-OP55X 820 —
GPT-6A 820 —
GROK-46H 820 -5.2%
GLM-5 784 -8.4%
KIMI-K3X 742 -8.4%
CL-FAB5H 742 -5.7%
CL-OP5H 718 -6%
CL-OP5X 708 -18.2%
CL-OP46H 696 -6.2%
CL-OP47H 688 -6.1%
GEM-38FH 677 +0.1%
GEM-37FH 655 -24.3%
GPT-56S 619 —
GPT-55H 580 —
CL-OP47 579 -0.7%
INKL 531 —
GEM-31P 512 —
GEM-3P 498 —
CL-OP46 496 —
CL-OP48 489 -0.2%

Live Feed

4mo ago model

GPT-5.5 Costs 49-92% More in Practice — OpenRouter's Switcher Cohort Study Breaks Down the Real Price

OpenRouter analyzed users who moved from GPT-5.4 to GPT-5.5 after launch. Net cost increases run from 49% on long-context work to 92% on short prompts. The model's verbosity reduction only kicks in above 10K input tokens.

4mo ago funding

OpenAI and Anthropic Both Launch PE-Backed Deployment Arms on the Same Day — $11.5B Combined

OpenAI finalizes a $10B joint venture with 19 PE investors. Anthropic closes a $1.5B deal with Blackstone, Goldman, and Hellman & Friedman. Both labs reach the same conclusion simultaneously: the bottleneck is no longer the model.

4mo ago funding

Sierra Raises $950M at $15B — Enterprise AI Customer Agents Hit Their Valuation Ceiling Test

Bret Taylor's Sierra closed a $950M round led by Tiger Global and GV, pushing valuation above $15B. The company builds AI agents that handle customer interactions end-to-end for brands including Weight Watchers, Sonos, and ADT.

4mo ago policy

Trump White House Eyes Pre-Release AI Model Vetting — Mythos Triggered the U-Turn

The administration that rolled back Biden's AI reporting rules is now considering mandatory government review of frontier models before public release. The trigger is Anthropic's Mythos, which can chain complex cyberattacks faster than human security teams.

4mo ago release

Ant Group Open-Sources Ling-2.6-1T: 1 Trillion Parameters, Fast-Thinking Architecture, Near GPT-5.4 on Execution Tasks

The Bailing model team's flagship open-weight model ships to Hugging Face and ModelScope with a hybrid MLA+LinearAttention design that cuts token waste in agentic workflows while matching GPT-5.4 on non-reasoning benchmarks. Free API trial on OpenRouter, extended one week.

4mo ago funding

Amazon Earned $30.3B in Q1 — $16.8B of It Was a Paper Gain on Anthropic

AWS grew 28% to $37.6B, its fastest quarter in 15 years. But strip the $16.8B pre-tax Anthropic stake revaluation and net income was closer to $13.5B. The backlog tells a more durable story: $364B in committed AWS spend, before counting the new $100B+ Anthropic deal.

4mo ago release

Microsoft Agent 365 GA: $15/Month to Govern Every AI Agent in Your Enterprise

Microsoft Agent 365 went generally available May 1 as a standalone $15/user/month control plane, or bundled into the new M365 E7 Frontier Suite at $99/month. It extends Defender, Entra, and Purview to non-human entities and ships alongside Copilot Cowork, co-built with Anthropic.

4mo ago release

1X Opens America's First Vertically Integrated Humanoid Factory: 10,000 NEO Units in Year One

OpenAI-backed 1X Technologies opened a 58,000-square-foot factory in Hayward, California on April 30, targeting 10,000 NEO home robots in year one and 100,000 by end of 2027. CEO Bernt Bornich: it is a manufacturing problem, not a robotics problem.

4mo ago release

Meta Acquires ARI to Build the Android of Humanoid Robots

Meta bought Assured Robot Intelligence on May 1, folding whole-body control models and a novel tactile sensor into Superintelligence Labs. The stated strategy: provide the AI layer, let hardware makers build the machines.

4mo ago research

LLMs Notice Hints 99% of the Time, Disclose Them 21%: Adobe Study Breaks Chain-of-Thought Safety Monitoring

A 9,154-trial study of 11 major LLMs finds models perceive contextual hints in 99.4% of cases but mention them in only 20.7% of chain-of-thought responses. The 78.7-point gap persists even when models are told they are being monitored, undermining CoT as a reliable safety oversight tool.

4mo ago release

Figure F.03 Walks From Factory Floor to HQ on Camera Only — Zero-Shot Sim-to-Real Stair Policy Ships at Scale

Figure AI's F.03 humanoid now navigates stairs autonomously using only onboard cameras, no LiDAR, no pre-mapped floors. The full locomotion policy was trained end-to-end with reinforcement learning in simulation and transferred zero-shot to physical hardware.

4mo ago research

OpenAI o1 Outperforms ER Doctors at 67.1% Diagnostic Accuracy in Harvard Science Study

A peer-reviewed study in Science from Harvard Medical School and Beth Israel Deaconess found o1-preview correctly diagnosed 67.1% of 76 real ER cases, against 55.3% and 50.0% for two expert attending physicians. Blinded reviewers could not distinguish AI diagnoses from human ones.

4mo ago policy

APRA Warns Frontier AI Could Arm Bank Attackers. Xero Just Signed a Multi-Year Anthropic Deal.

Australia's banking regulator says frontier AI models could equip attackers with capabilities banks are ill-prepared to counter. That same week, Xero locked in a multi-year partnership with Anthropic to automate accounting workflows with Claude.

4mo ago policy

Academy Bans AI From Acting and Writing Oscars: Human Authorship Now a Formal Eligibility Rule

The Academy of Motion Picture Arts and Sciences has updated its awards rules to bar AI-generated performances and AI-authored scripts from Oscar eligibility. Alongside the change, a decades-old limit on double acting nominations was lifted.

4mo ago release

Augment Code Ships Prism: Per-Turn Model Routing Cuts Coding Agent Costs 20-30% at Frontier Quality

Augment Code has launched Prism, a model router that makes routing decisions at the individual turn level rather than the session level, matching quality against GPT-5.5 and Opus 4.7 at 20-30% lower cost. Teams sending 10,000 messages per month can expect to save $20,000.