Muse Spark 1.1 Replaces Qwen3.7-Max in helloai's Frontier Set
Meta ships Muse Spark 1.1 with a public API at $1.25/$4.25. helloai drops Qwen3.7-Max and adds Meta as its budget agentic pick at 1487 Elo.
On July 9, Meta Superintelligence Labs shipped Muse Spark 1.1 and opened the Meta Model API in public preview — the first time developers can buy Meta inference directly. helloai replaces Qwen3.7-Max with Muse Spark 1.1 in its six-model frontier set: $1.25/$4.25 per million input/output tokens, 1M context, and LMArena text overall at 1487 Elo (preliminary, 13,591 votes). The admission drops Alibaba's API flagship to make room for Meta under helloai's six-model cap; Qwen3 32B remains in the open-weight section as a separate track.
The positioning overlap is why the swap makes sense. Qwen3.7-Max entered helloai in June as the budget agentic option — 80.4% on SWE-bench Verified, $2.50/$7.50, 1475 Elo. Muse Spark 1.1 targets the same procurement story at lower headline rates and higher arena rank: Meta claims MCP Atlas 88.1, JobBench 54.7 against Opus 4.8's 48.4, and native multi-agent orchestration with MCP servers and computer use. Coding benchmarks are mixed — SWE-Bench Pro 61.5 trails Opus 4.8's 69.2 — but agentic tool chains are where Meta invested.
The business shift behind the model matters for readers who built on Llama. Muse Spark is closed and metered; Meta's free-weights era is not extended to this line. Zuckerberg's launch-day post framed Spark as strong agentic performance at a very low price — competitive with Grok 4.5 and well below Opus and GPT-5.6 Sol on output tokens. US developers get $20 in free API credits; EU and global API access timelines are not fully documented at launch. Consumer access is free via meta.ai Thinking mode.
For helloai's /api/recommend routing, the competitive stack reorders slightly. Muse Spark 1.1 slots fifth by Elo at 1487, above GPT-5.6 Sol at 1484 and below Grok 4.5 at 1490. It becomes the cheapest tracked frontier API on input and the strongest Meta representation helloai has ever carried — previously rejected in May when no public API existed. Teams that routed coding tasks to Qwen3.7-Max for cost should retest Muse Spark on their agent harnesses; benchmark claims are vendor-reported and eval-hygiene questions surfaced on launch day.
helloai's admission rules require a public API, proven provider, 200K+ context, and arena signal within 25 points of the set floor. Muse Spark passes on family Elo; 1.1-specific votes are hours old, so curated 1487 stands until the weekly Elo scraper matches arena aliases. The May 31 rejection of muse-spark is superseded. Meta joins Anthropic, Google, xAI, and OpenAI in the tracked frontier set — six labs, six models, one budget-agentic slot, now painted Meta blue.