Tech • AI • Robotics • Game
Meta M Wonder flops, Base 44 wins, Lucy raises $20M
Briefing
Meta models miss coding hype
Meta's M Wonder and M Wonderful arrived with benchmark credentials suggesting a low-cost coding challenger, including a roughly 56 score on Artificial Analysis. But hands-on coding and simulation trials reportedly fell well short of those expectations. In practical use, both models lagged behind Claude Opus 5.5 and often behind Gemini 4 Argon. The gap underscored a widening split between leaderboard optics and real-world coding reliability.
M Wonderful stumbles in simulation
A crash-simulation test exposed the sharpest weakness in M Wonderful. Instead of rendering a believable impact, the model generated erratic behavior, including attempts to avoid the wall and visibly implausible motion. That failure mattered because simulation tasks probe code quality, physics logic and debugging resilience at once. The result suggested that benchmark strength did not translate into robust agentic coding performance.
Naming muddles Meta's lineup
Coverage of M Wonder versus M Wonderful highlighted a branding problem as much as a product one. The two models appeared close in name but different in capability, with M Wonderful shown near third place on one M Code leaderboard while the cheaper variant carried more of the value narrative. That blurred distinction risks confusing buyers comparing price, access and expected coding quality. For enterprise evaluation, the episode was a reminder that model family labels can obscure meaningful performance differences.
Base 44 tops builder showdown
Base 44 ranked first in a five-way comparison of AI website builders aimed at freelance client delivery. It outperformed Lovable, Replit, Bolt and Rocket across identical tasks measuring speed, functional completeness, design quality and operating cost. The strongest differentiator was consistency: Base 44 finished complex jobs quickly with few or no repairs. That made it the clearest fit for agencies and solo operators selling fixed-scope builds.
Base 44 wins on speed
In a local business website brief for Sparkle Wash, Base 44 completed the build in about 3 minutes and scored 24/25. Lovable followed at 21, ahead of Bolt at 19, Replit at 18 and Rocket at 15. The top output included the expected commercial sections and needed minimal cleanup before client presentation. For routine brochure sites, rapid execution and polish proved more valuable than feature breadth alone.
AI app test exposes gaps
The hardest assignment was an AI-powered website audit app requiring URL analysis, generated findings, authentication and user history. Base 44 again led with 24/25, delivering a working app in roughly 4 minutes without fixes. Lovable reached 18 only after more than 20 minutes and multiple corrections, while other rivals fell further behind. The result showed that agentic builders still diverge sharply once workflows move beyond landing pages into real software behavior.
Lucy scales biomarker platform
Lucy, a European preventive health startup, is expanding after raising $6.5 million in seed financing and $20 million in Series A funding. Its platform combines blood, urine and saliva testing, and sometimes microbiome analysis, across roughly 100 biomarkers. Software and AI-assisted interpretation are then reviewed by medical staff to produce personalized recommendations on sleep, nutrition, exercise and supplements. The company is positioning the service as proactive wellness infrastructure rather than traditional diagnosis.
Lucy pushes Europe expansion
Lucy says early detections have already produced high-stakes outcomes, including one case that led to immediate surgery and a later claim that the intervention likely saved a life. That narrative strengthens the commercial case for consumer-paid monitoring despite Europe's more conservative healthcare culture. The company is now active in France, the UK, Ireland and Portugal, with broader regional rollout planned. Its bet is that wellness-first framing can unlock demand before symptoms appear and before public systems intervene.
Videos covered
Muse Wonder : le nouveau modèle de Meta DÉTRUIT Claude et GPT-6 ?
- •Benchmark scores raised expectations
- •Confusion between M Wonder and M Wonderful
- •Crash simulation exposed major weaknesses
Best AI Tool to Build Local Websites & Sell Them
- •Five builders tested across identical client tasks
- •Local business website favored speed and polish
- •AI web app exposed major capability gaps
Comment il a levé 26,5 M€ en moins de 2 ans pour sa startup dans la santé
- •A preventive health platform built on biomarkers
- •Early detection has already produced life-saving cases
- •Funding and expansion strategy