AI Model Timeline

Tracking the accelerating release frequency of frontier AI models.

Muse Ember

Meta·

LLMssmallcloud

Overview

Meta's smallest and lowest-priced reasoning model, a new tier below Muse Spark aimed at high-volume agent work: tool calls, coding sub-agents, computer use and document workflows. Available through the Meta Model API (model id muse-ember-1.0), all Muse Code plans and Google Cloud. 1M-token context with up to 512K output tokens. List price $0.10 / $0.50 per MTok up to 200K prompt tokens and $0.30 / $1.50 above, cached input $0.004. Meta publishes no statement on weights; Artificial Analysis lists the model as proprietary. Meta's launch table carries none of the tracked academic benchmarks. Its own figures are OSWorld 2.1 69.5 (partial credit, internal harness), Agents' Last Exam 61.6 and its internal Personal Superintelligence Index 46.8, none of them tracked here. Artificial Analysis Intelligence Index v5.0.1: 36 at max effort.

Capabilities and innovations

Reasoning with selectable effort (minimal to max)Tool calling and agentic workflows (Responses API)Computer use1M-token context, up to 512K output tokensText, image, audio and video input (per Meta; Artificial Analysis lists text and image)First small, low-cost tier in the Muse lineTiered pricing by prompt length (200K threshold)

Benchmarks

BenchmarkScoreSource
Terminal-Bench 4.039.39%independent
Terminal-Bench 4.0 39.39% (±2.31), measured by Vals AI with the mini-swe-agent harness at max effort, rank 17/64 (accessed 2026-11-13). Vals AI is the reference harness for this field. 4.0 is its own generation and not comparable with the 2.x or 3.0 fields.
  • GPQA Diamond: not reported by the vendor — Not in Meta's launch table; no standalone value on Vals AI or Artificial Analysis (checked 2026-11-13).
  • Humanity’s Last Exam (no tools): not reported by the vendor — HLE-Diamond appears only as a component of the Artificial Analysis index, a different subset and without a standalone Ember value (checked 2026-11-13).
  • SWE-bench Verified: not reported by the vendor
  • SWE-bench Pro: not reported by the vendor — Meta reports FrontierCode 2.0-AA (27.2) and the Coding Agent Index-AA (46, run inside Muse Code) instead; neither is a tracked field.
  • MMLU: not reported by the vendor
  • MMLU-Pro: not reported by the vendor
  • Terminal-Bench 2.x: not reported by the vendor — Only Terminal-Bench 4.0 is reported, recorded in terminal_bench_4.
  • Terminal-Bench 3.0: not reported by the vendor
  • DeepSWE v1.1: not reported by the vendor — No DeepSWE figure in any version on Meta's launch pages (checked 2026-11-13).
  • MMMU-Pro: not reported by the vendor

API pricing

$0.1 per million input tokens, $0.5 per million output tokens (USD, cheapest OpenRouter endpoint, checked 2026-11-13)

Links

More from Meta

Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.