OpenRouter now lists Meta’s Muse Spark 1.1 at the same aggressive token price as Meta’s new developer API, adding interface compatibility without adding another inference provider. Early cache and reliability data make the offer easier to assess, but not yet proven for sustained agent workloads.
Meta’s Muse Spark 1.1 is now available through OpenRouter, one week after Meta introduced the multimodal agent model and opened its own Model API in public preview. The listing broadens how developers can reach the model. It does not broaden who supplies the inference.
That distinction is the center of the offer. Developers gain a familiar gateway and several request formats, while Meta remains the only upstream provider. The more consequential test is therefore not whether OpenRouter can list Muse Spark, but whether Meta’s price, caching and service quality make one hosted endpoint dependable enough for long-running agent work.
Meta announced Muse Spark 1.1 on July 9 and said developers could access it through the new Meta Model API for the first time. It also made the model available in “Thinking” mode in the Meta AI app and on meta.ai. OpenRouter’s model page records a July 16 release on its service.
Through OpenRouter, developers can call the model with an OpenAI-compatible chat-completions endpoint, an OpenAI Responses endpoint or an Anthropic-style Messages endpoint. The listing says Muse Spark accepts text, images, video, audio and PDFs, returns text and has a one-million-token context window. It also advertises structured output, parallel function calling, built-in search with citations and adjustable reasoning effort.
Those are distribution and integration gains. They are not a new market for hosting Muse Spark. OpenRouter lists one provider, Meta, and says it forwards every request directly to that provider with no routing decision. OpenRouter’s general recovery mechanism can move a failed request to another healthy provider when one is available and the request permits it; for Muse Spark 1.1, the page shows no alternative endpoint to select.
The posted rate is $1.25 per million input tokens and $4.25 per million output tokens. A launch report, drawing the price from Reuters, described the rate as competitive but slightly above Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.6 Luna. That comparison places Muse Spark among lower-cost agent models; it does not establish that it is the cheapest option for every workload.
OpenRouter’s observed pricing gives the list rate more context. Its effective-pricing panel showed a weighted average input price of $0.818 per million tokens, an output price of $4.25 and a 39.2% cache-hit rate, with all token share going to Meta. OpenRouter says the panel reflects what customers paid after prompt caching. Because the model had only just appeared on the service, those figures are an early snapshot rather than a settled cost curve.
The same caution applies to operations data. Across the three-day period displayed on the archived page, Meta’s endpoint averaged:
| Measure | Reported average |
|---|---|
| Throughput | 114 tokens per second |
| Latency | 4.38 seconds |
| End-to-end latency | 10.03 seconds |
| Tool-call error rate | 3.27% |
| Structured-output error rate | 10.26% |
| Provider uptime | 100% |
The uptime figure limits a simple “single provider equals unreliable” conclusion. The error metrics limit the opposite conclusion. They measure different aspects of service behavior, and three days cannot show how the endpoint performs through sustained load, long sessions or a production incident. Nor do benchmark scores reveal how many paid retries a particular agent workflow will require.
Meta presents Muse Spark 1.1 as more than a coding assistant. The company says it can plan as a main agent, delegate work across parallel subagents, operate as a subagent that knows when to escalate, and manage its one-million-token context through retrieval and compaction. It also says the model can decide when scripting is faster than operating an interface directly.
For coding, Meta claims improvements on complex bugs, new features, large migrations and workflows that combine screenshots with code changes. Its public evidence is still largely company-selected: Meta says Muse Spark 1.1 improved substantially over the first Muse Spark on an internal coding evaluation and was competitive with leading alternatives. The announcement does not publish that internal test’s results in the archived body.
OpenRouter displays externally sourced benchmark results with defined, but broad, comparison sets. Artificial Analysis gave the model a 71.3 Coding Index score, better than 89% of models compared on OpenRouter; a 50.6 Intelligence Index score, better than 86%; and a 37.5 Agentic Index score, better than 82%. These rankings support the case that Muse Spark is competitive. They do not isolate task-completion cost, provider resilience or performance inside a developer’s own tool stack.
Meta was also not first to this category. The launch report notes that OpenAI and Anthropic already offered comparable agentic models, while the original Muse Spark appeared in April. Muse Spark 1.1 is a stronger and more accessible entry into an existing contest, not the creation of a new one.
The model’s safety claims require similar boundaries. Meta says its own pre-deployment evaluations found Muse Spark 1.1 within safe margins for chemical and biological risk, cybersecurity and loss of control, and more resistant to jailbreaks and prompt injection. OpenRouter’s telemetry evaluates endpoint behavior; it does not independently confirm those frontier-risk conclusions.
OpenRouter distribution matters partly because Meta is trying to turn costly AI infrastructure into a direct developer business. A July 9 analysis of Meta’s revenue plan described Muse Spark 1.1 as the first time Meta had made its models publicly available to developers through an API and the first time it would charge for that access.
The same analysis reported that Meta was on track to spend more than $200 billion over two years on its AI buildout, including data centers, chips and infrastructure, with another $600 billion committed through 2028. Those totals cover a much larger program than Muse Spark 1.1. They explain why a low-priced API is strategically significant without proving that this model can recover those costs or win share.
OpenRouter gives Meta another channel through which developers can compare Muse Spark with substitutes using a shared interface. But because the underlying endpoint and price still come from Meta, the listing redistributes access more than control.
The unresolved question is whether Muse Spark’s apparent value survives an entire agent loop. The useful comparison is not a benchmark score or one million tokens in isolation. It is the cost and time required to finish an accepted task, including reasoning tokens, cache reuse, tool failures, malformed structured responses and retries.
Developers also need a longer operating record: latency under sustained demand, error rates across different tools and schemas, and evidence of how the sole endpoint behaves during a disruption. A second provider would add a direct test of hosting economics and resilience. Until either that alternative appears or Meta’s endpoint builds a longer record, OpenRouter has made Muse Spark easier to buy and compare—not less dependent on Meta.
OpenAI has put conversations and Projects back in its redesigned ChatGPT desktop app and enabled cloud Work threads to move across devices, correcting the launch's biggest usability failures without merging local Work or Codex histories.
Meta reportedly plans to put departing AWS compute executive Dave Brown to work on its data-center buildout, adding hyperscale operating experience while leaving any customer-facing cloud business conditional and undefined.
China has paired a five-year AI training offer for developing countries with cooperation centers, a weather-warning rollout and a new 29-country organization. The package gives Beijing a platform for influence, but no budget, selection rules or delivery timetable has been published.
RoboTTT extends a robot policy’s visuomotor context to 8,000 timesteps and raises its average rubric score from 42% to 79%, but the evidence comes from three author-run assembly tasks and the longest task was completed in only two of ten trials.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
Nebius has arranged its first senior secured facility against an operating GPU deployment and one customer’s cash flows. The deal adds project-level debt to its expansion toolkit, but an unnamed customer and a larger rival financing limit what it proves about the rest of Nebius’s backlog.
Cursor says automatically running a repository-root git.exe on Windows does not meet its criteria for patching, while the researcher calls it an untrusted-search-path defect and separate research shows the same weakness across several competing AI coding tools.
Huawei publicly displayed a 16-cabinet Atlas 950 configuration rated at 1 EFLOPS in FP8, providing tangible evidence of its system-scale AI strategy while leaving price, power use and sustained workload performance undisclosed ahead of the full system's planned fourth-quarter release.
CIA Director John Ratcliffe said US intelligence is consistent with an estimate that Russian recruits last 20 to 30 minutes on Ukraine’s battlefield, but the public trail leads to an unsourced claim about assault troops and does not establish a representative average.
UK testing places leading open-weight models four to seven months behind selected closed-model cyber results, yet longer attack chains, U.S. benchmarks and mixed cost comparisons show why that interval is a warning signal rather than a universal capability clock.
TSMC reached the top of its second-quarter revenue guidance and raised its 2026 outlook, but a one-off investment gain boosted profit growth while the company committed more capital to 2-nanometer production, advanced packaging and an undated Arizona expansion.
Moonshot AI's Kimi K3 added to a global technology selloff with near-frontier performance, but unreleased weights, mixed cost comparisons and a recommended 64-accelerator deployment leave its effect on chip demand unresolved.
Moonshot AI has made Kimi K3 available through its apps and API, pairing a 2.8-trillion-parameter architecture with early frontier-level results, but the model's open-weight claim cannot be tested until its weights and technical report arrive.
Two binding EU decisions require Google to give rival AI services comparable access to 11 Android features and offer eligible search competitors a restricted, anonymized dataset, but phased deadlines, certification, pricing and privacy safeguards leave the competitive effect unproven.
Databricks has signed a term sheet for a Coatue-led financing at a $188 billion valuation, while unidentified sources put the round at $3 billion. The proposed capital would deepen its push into AI governance, data agents and operational databases, but the transaction remains open and the company supplied no new operating figures.
Netflix says generative AI workflows were used on roughly 300 titles in 2026, but its only quantified example tied to that disclosure covers 17 minutes and does not establish how much finished material, spending or labor changed across the slate.
Gemini 3.5 Pro missed its expected June rollout. An anonymous-source account says a late-June data change fell short of Google's coding goals, but Google has confirmed only partner testing—not the reported cause, a new date, or public results and pricing.
Google is bringing Gemini Omni editing and a reusable face-and-voice avatar to Vids, but the sharper distinction is account-level identity across Vids and Gemini rather than a new category of AI video; Vids already offered Veo generation and customizable avatars, while specialist rivals already sell digital presenters.
Intel plans to adopt Gemini Enterprise across engineering, supply-chain and corporate work while extending chip-simulation capacity into Google Cloud, but the companies supplied no rollout schedule, contract value or performance baseline.
The Navy has approved an immediate department-wide framework for turning data into operational effects, but the public rollout leaves budgets, owners, deadlines, performance measures and assurance rules to a promised implementation roadmap.