Meta Says Its New Cheapest AI Model Can Beat GPT-6 Luna

Meta has introduced Muse Ember, its smallest and most efficient AI model, claiming it delivers better performance per dollar than OpenAI's GPT-6 Luna and Anthropic's Claude Haiku 5.5 throughout a number of reasoning and agent-based tasks.

Key points

• Muse Ember scored 40.9% on Terminal-Bench 4.0, ahead of Claude Haiku 5.5 (39.2%) and GPT-6 Luna (16.4%).
• At maximum reasoning effort, Muse Ember scored 36 on the Artificial Analysis Intelligence Index, matching Claude Haiku 5.5 and exceeding GPT-6 Luna's score of 28.
• Nevertheless, Claude Haiku 5.5 performed better on FrontierCode 2.0-AA, scoring 34.9% compared with Muse Ember's 27.2%.
• According to Meta, Muse Ember costs approximately $0.14 per Artificial Analysis task, compared with $2.39 for Muse Spark 1.3.
• Despite the difference in cost, the two models scored 36 and 37, respectively, on the Artificial Analysis Intelligence Index.

Notably, the new model focuses on complex reasoning and repetitive AI agent operations, including coding, tool calls, task routing, information extraction, and processing lengthy documents.

While outperforming GPT-6 Luna throughout the tested reasoning settings, according to Meta, Muse Ember delivers performance comparable to Claude Haiku 5.5 at roughly one-third the cost in certain configurations.

Meta notes Muse Ember sits on the efficiency frontier of the Artificial Analysis leaderboard, meaning no tested model achieves a higher intelligence score at a lower cost per task.

Meanwhile, the model additionally performed well in agent-based tasks, scoring 37.8% on AutomationBench-AA, compared with 20.2% for GPT-6 Luna and 13.9% for Claude Haiku 5.5.

Notably, the after table compares the models throughout major benchmarks at their maximum reasoning settings.

Meta's Muse AI Now Lets You Create an App Within an App. Benchmark Muse Ember GPT-6 Luna Claude Haiku 5.5. Artificial Analysis Intelligence Index 36 28 36. AutomationBench-AA 37.8% 20.2% 13.9%. Agents' Last Exam 61.6% 57.1% 57.0%. Personal Superintelligence Index 46.8% 37.0% 36.1%. Terminal-Bench 4.0 40.9% 16.4% 39.2%. Coding Agent Index-AA 46 31 37. FrontierCode 2.0-AA 27.2% 25.2% 34.9%. GDP.pdf-AA 22.8% 22.8% 20.8%. OSWorld 2.1 69.5% 50.5% 68.9%.

PerceptionBench 80.6% 62.8% 78.6%.

Bold figures indicate the highest score in each benchmark. Results come from Artificial Analysis and Meta's internal evaluations. Some tests employ different evaluation setups.

Muse Ember supports text, images, audio, and video, allowing it to analyze screenshots, charts, scanned documents, recordings, and other content.

For context, the model features a 1-million-token context window and supports outputs of up to 512, 000 tokens, allowing it to process sizeable codebases and document collections.

It can additionally perform computer-based tasks, review contracts, analyze documents, run coding tests, and complete workflows involving multiple connected applications.

Meta notes developers can pair Muse Ember with Muse Wonder, using the larger model for complex planning while Ember handles repetitive tasks. The after table summarizes its specifications. Model name Muse Ember 1.0. Context window 1 million tokens. Maximum output 512, 000 tokens. Backed inputs Text, images, audio, video. Main applications AI agents, coding, tool calls, document processing. API backing Responses API. Availability Muse Code, Meta Model API. Cloud platforms Google Cloud, Oracle Cloud.

Muse Ember is available through Muse Code and the Meta Model API under the name muse-ember-1.0. Meta has confirmed the after pricing. Input tokens (1 million) $0.10. Cached input tokens (1 million) $0.004. Output tokens (1 million) $0.50. Estimated cost per Artificial Analysis task $0.14.

Despite the difference in cost, the two models scored 36 and 37, respectively, on the Artificial Analysis Intelligence Index. Muse Ember vs. Muse Spark 1.3.

While Muse Spark 1.3 targets more demanding coding and agentic workflows, muse Ember is designed for repetitive coding and agent-based tasks. Feature Muse Ember Muse Spark 1.3. Artificial Analysis Intelligence Index 36 37. Cost per task $0.14 $2.39. Input cost (1M tokens) $0.10 $1.25. Output cost (1M tokens) $0.50 $4.25.

Notably, the figures show that Muse Ember achieves a similar intelligence score at a substantially lower estimated cost per task. Anthropic Building Tool to Expose Chinese Models Secretly Using Claude. Safety and Future Intends.

Meta notes it evaluated Muse Ember under its Superintelligence Scaling Framework, covering cybersecurity, chemical and biological risks, and potential loss-of-control scenarios.

According to the firm, the model remained below its high-risk thresholds in the categories tested.

Meta additionally assessed whether Muse Ember follows user instructions, protects sensitive information, resists malicious instructions embedded in external content, and requests human approval before high-impact actions.

Muse Ember is the first model Meta trained using Muse Wonder Full. It joins Muse Spark and Muse Wonder in the company's AI model lineup.

Meta notes all three models are now available on Google Cloud and Oracle Cloud, with further efficiency-focused models at present in development. Stay Connected with ProPakistani.

Obtain the latest tech news, telecom insights, and product launches wherever you prefer. Follow on Google Discover. Follow on Google News Join WhatsApp. See more ProPakistani stories in Google Search and Top Stories.

Technology and Automotive Specialist covering the latest cars, smartphones, AI breakthroughs, and.

For now, meta Says Its New Cheapest AI Model Can Beat GPT remains the part of the story worth watching, and further updates are likely as more details are confirmed.