Search “Meta”

Matched fieldsTitle、Summary、Body

A new Meta paper finds that reinforcement learning post-training makes agents perform better on first attempts, but fewer tasks can be solved within up to 128 attempts. The study tested 14 sets of base models and post-trained models on 3 agent benchmarks, and built the PTGS sampler, which raises temperature for difficult prompts and lowers it for easy prompts. On WebShop, the gemma-4-31B base model solved over 85% of tasks within 128 attempts, while the post-trained version solved 56%; the autho

Matched fieldsTitle、Summary、Body

A new paper from Princeton and Meta proposes the Self-Direction method, which lets a failed agent reflect on the current run and trains the lessons into a LoRA adapter about 0.02% the size of the model; a router then loads the closest adapter when a similar failure is encountered. On Qwen3.6-27B, the GAIA task success rate rises from 38.8% to 55.8%, outperforming in-context reflection and the Skills.md file approach. Of 185 errors that had occurred previously, 67.3% of fixes were still effective

Matched fieldsTitle、Summary、Body

Meta has quietly launched Style Sketch, an AI fashion app that can connect to Instagram to build a virtual wardrobe from already-posted photos, and also supports manually adding items by taking photos, recording videos, or uploading images from the camera roll. The app recommends outfits based on the user's style and lets users try them on a "realistic model" that matches their body type and skin tone, and it also supports generating day-by-day outfit plans based on destination, trip length, and

Matched fieldsTitle、Summary、Body

Meta has published the production deployment results of Catching JiTTests, a system that uses LLMs to automatically generate executable tests for pending code changes, exposing defects before the code is merged. The system employs two workflows, intent-aware and dodgy diff, and uses an LLM-as-Judge ensemble together with a rule-based method, RubFake, to filter false positives, while reducing manual line-by-line review through a human-on-the-loop approach. Meta says 589 engineers have participate

Matched fieldsTitle、Summary、Body

ARC Prize published the validation results of Meta's Muse Spark Open on ARC-AGI: ARC-AGI-1 at 89.0% with $0.29 per task, ARC-AGI-2 at 57.5% with $0.73 per task, and ARC-AGI-3 at 0.1% under the standard harness with $3.9K. ARC Prize states that its v1/2 scores are close to those of Haiku 5.5 and GPT-6 Luna, but the cost is about 9 times that of the latter; the model is priced on OpenRouter at $1.25 per million input tokens and $4.25 per million output tokens. As the reasoning tier increases, ARC-

Matched fieldsTitle、Summary、Body

ARC Prize tested Meta Spark Open on ARC-AGI: it scored 89.0% on ARC-AGI-1, 57.5% on ARC-AGI-2, and only 0.4% on ARC-AGI-3. Its v1/v2 scores are close to those of Haiku 5.5 and GPT-6 Luna, but its cost is about 9 times theirs. Across providers on OpenRouter, the model costs $1.25/M for input and $4.25/M for output, while Haiku 5.5 and GPT-6 Luna both cost $0.10/M for input and $0.50/M for output. As the reasoning tier increases, ARC-AGI-2 rises from 29.4% at medium (averaging 55k tokens/task) to

Matched fieldsTitle、Summary、Body

ARC Prize announced the validation results of Meta Muse Spark Open on ARC-AGI: ARC-AGI-1 is 89.0% at $0.29 per task, ARC-AGI-2 is 57.5% at $0.73 per task, and ARC-AGI-3 is 0.4% at $3.0K under the provider adapter harness and 0.1% at $3.9K under the standard harness. ARC Prize states that its scores on v1/v2 are similar to those of Haiku 5.5 and GPT-6 Luna, but the cost is about 9 times that of the latter; the model is priced on OpenRouter at $1.25 per million input tokens and $4.25 per million o

Matched fieldsTitle、Summary、Body

Meta announced support for research at five universities—Caltech, Ohio State Med, LSU, Indiana Univ, and UTHealth Houston—opening its latest Muse models (Muse Wonder, Muse Spark), open-weight models Muse Glimmer and Muse Spark Open, computer vision models such as SAM, V-JEPA, and DINO, and the multilingual model SeamlessM4T to researchers. These universities will use the models for directions including mathematical and physical science research, X-ray screening and heart failure detection, infan

Matched fieldsTitle、Summary

Meta announced support for research at five universities—Caltech, Ohio State Med, LSU, Indiana University, and UTHealth Houston—opening Muse Wonder, Muse Spark, the open-weight models Muse Glimmer and Muse Spark Open, as well as SAM, V-JEPA, DINO, and SeamlessM4T to researchers. Caltech will test whether large language models can accelerate research in mathematics and physics, while UTHealth Houston will use Muse Glimmer, one of the smallest open-weight models, to improve brain cancer treatment

Matched fieldsTitle、Summary

Matched fieldsTitle、Summary、Body

Danish optics company NIL Technology has released the ultra-compact eye-tracking camera metaEye, which it calls the world's first camera designed entirely with meta-optics, with overall dimensions smaller than a grain of rice. The product integrates two metalens optical elements (MOE), a bandpass filter, and multiple optical apertures, and is mounted directly on the image sensor's cover glass, eliminating the need for traditional refractive lenses. NILT says relevant customers are moving toward

Matched fieldsTitle、Summary、Body

Nvidia has invested $3 billion in Meta's Muse agent and the Muse series models, and the two sides announced that the Muse models (Wonder and Wonder Full) and the agent will run on Nvidia infrastructure. Meta plans to expand its deployment of Nvidia hardware, including using Vera Rubin and Vera CPU for next-generation AI agents, and the two sides are also exploring using Nvidia's OpenShell reference architecture for agent security. Meta's consumer-grade Muse agent has reached 10.9 million downloa

Matched fieldsTitle、Summary、Body

On October 26, Meta announced it would raise $97.5 billion through an equity offering for AI infrastructure; Nvidia announced a $3 billion investment the same day, and AMD announced a $1 billion investment the following day. Meta plans to deploy the Nvidia Vera Rubin platform, Vera CPU, Spectrum-X Ethernet, and BlueField-4 DPU, and to deepen its collaboration with Nvidia on the Muse model and agents; on the AMD side, Meta has installed and tested the first batch of Helios rack-scale systems and

Matched fieldsTitle、Summary、Body

Meta is partnering with five U.S. universities—Caltech, Ohio State University, Louisiana State University, Indiana University, and UTHealth Houston—providing $2 million in funding and access to its latest models to support research in medicine, physics, chemistry, biology, computer science, and mathematics using its models. Specific projects include using the open-weight SAM model to monitor tracheostomy tube displacement in infants, using multimodal AI to help dementia patients communicate, and

Matched fieldsTitle、Summary、Body

Meta updated its commercial terms this month to allow it to use content from associated domains, web pages, or URLs to develop, improve, research, train, and maintain Meta products, sparking discontent among artists and small business owners who rely on Instagram and Facebook. The effective date was postponed from October 30 to November 20, and Meta subsequently removed the word "train" from the list of permitted uses and limited the uses to "advertising and business products." UK and EU users c

Matched fieldsTitle、Summary、Body

Meta updated its terms for business accounts to allow it to use content from linked websites and pages to develop, improve, and train AI, sparking dissatisfaction among small businesses and artists who use Instagram. UK and EU users can currently submit an objection form in the Instagram Privacy Center, and some users say they have received emails from Meta approving their objections. The terms were originally set to take effect on October 30 but were changed to November 20 after objections, and

Matched fieldsTitle、Summary、Body

ed India's largest-ever IPO, planning to issue up to 27 crore new shares, representing about 2.93% of post-issue share capital, with RIL's stake to be diluted from 66.43% to about 64.4%. Meta affiliate Jaadhu Holdings holds a 9.98% stake, and Google International holds 7.73%. As of June, the company had cumulatively filed 7,220 patent applications, of which 1,043 were granted; in FY26, 11 subsidiaries were loss-making, rising to 15 in Q1 FY27, with a combined loss after tax of 20.5 crore rupees.

Matched fieldsTitle、Summary、Body

NVIDIA announced a $3 billion investment in Meta's equity offering, and the two companies will expand their collaboration on the personal agent product Muse and the Muse series of models to improve model inference and training performance. The investment is part of Meta's $95 billion fundraising plan, and AMD and Broadcom have also expressed interest in investing. Meta expects full-year capital expenditure of $135 billion to $145 billion and will consider adopting NVIDIA's OpenShell secure agent

Matched fieldsTitle、Summary、Body

Nvidia and Meta announced an expansion of their collaboration on Muse AI models and agents, with Nvidia set to invest $3 billion in Meta's equity offering. The two companies will collaborate around the Nvidia OpenShell reference architecture, including Nvidia OpenShell for secure deployment and Nvidia Sentry for agent behavior monitoring; Meta will also extend its use of Nvidia infrastructure to Vera Rubin, Vera CPU, as well as Spectrum-X and BlueField-4 Ethernet. Meta currently uses Nvidia infr

Matched fieldsTitle、Summary、Body