Muse Spark 1.3 vs Claude Opus 5: Benchmarks & Cost (2026)

Meta listed Muse Spark 1.3 on 2 September 2026, four weeks after 1.2. It is the first Muse Spark release to land within three points of a frontier Anthropic model on a neutral harness — and it does so at a quarter of the price, with video input that Claude Opus 5 does not accept at all.

The boring version of this comparison — "cheap model loses to expensive model" — stopped being true at 1.3. But the gap that remains sits in exactly the place buyers care about most. Every figure below names the source beside it, with vendor-reported and independently-run numbers kept apart, because they are not the same kind of evidence. Artificial Analysis rebased its Intelligence Index from v4.1.1 to v4.3.2 after this article first ran, so every score here was re-pulled on 5 October 2026 and is stamped with the index version it came from.

What do the two models actually give you?

Specifications from the live OpenRouter model API and Anthropic's own model documentation, both read on 3 September 2026.

Audio is missing from that list deliberately. Meta documents audio on Muse Spark 1.3 as "not fully supported," and both OpenRouter and Artificial Analysis have since dropped audio from 1.3's listed input modalities while still listing speech input on 1.2 — see the multimodal section below.

Three of those rows are structural rather than incremental: Muse Spark 1.3 is 4× cheaper on input and 5.9× cheaper on output, it takes video natively, and its output ceiling is 7.4× higher. Everything else is close enough to be a wash — the context windows differ by 4.9%, which will never decide a purchase.

Anthropic's documentation adds a detail that now outweighs most of the table above: as of October 2026 Claude Opus 5 is listed under "Legacy models (still available)". Anthropic's standing advice is to "start with Claude Opus 5.5 for most workloads," holding Claude Fable 5.1 back for "demanding reasoning and long-horizon agentic work." Opus 5.5 is cheaper than Opus 5 — $4.00 / $20.00 per million tokens against $5.00 / $25.00 — and scores 57.6 on Artificial Analysis's Intelligence Index v4.3.2 against Opus 5's 50.8. If you are picking a model today rather than auditing one you already run in production, the pairwise answer below is not the whole answer; see our Claude Opus 5.5 complete guide.

What do independent benchmarks say?

Artificial Analysis retired Intelligence Index v4.1.1 — the version that produced the 61-vs-63 figures this article originally carried — and replaced it with v4.3.2, which rebased the scale and swapped in new evaluations. The two versions are not comparable and there is no conversion factor between them. On v4.3.2, read 5 October 2026: Muse Spark 1.3 (Max) scores 48.1 against Claude Opus 5 (Max) at 50.8, ranking #23 of the 224 models in its comparison class against Opus 5's #16. The median for that class is 26, so both sit far above average. The context that matters: Muse Spark 1.2 scores 39.6 on the same v4.3.2 index. An 8.5-point generational gain in four weeks cut the gap to Opus 5 from eleven points to under three.

Terminal-Bench 4.0 — the verified agentic-coding leaderboard — tells the opposite story, and it is the single most important table in this article. One thing has changed since this piece first ran: Opus 5 no longer leads it. The board has grown, and the same 51.8% ± 3.4 Opus 5 posted in July now ranks tenth. Selected rows, all 330-trial runs, read 5 October 2026:

No Muse Spark model appears on that board at any rank. Not 1.1, not 1.2, not 1.3. The same holds on SWE-bench Verified, where the official submission archive contains no entry for either Muse Spark or Opus 5.

That is no longer the whole story, and the change works against Muse Spark. Artificial Analysis folded Terminal-Bench 4.0 into Intelligence Index v4.3.2 as one of its ten constituent evaluations, and runs it in-house against every model it tracks. So Muse Spark 1.3 does now have an independently-run agentic-coding number — it is simply a weak one: 33.3% against Opus 5's 49.0% on Artificial Analysis's own Terminal-Bench 4.0 pass. Muse Spark 1.2 managed 7.1% on the same evaluation, so 1.3 is a real leap; it is still 15.7 points behind a model Anthropic has already superseded, and 26.3 behind Opus 5.5's 59.6%. The argument has changed shape rather than direction — it used to be "one has evidence and one has none," and it is now "both have a number and Muse Spark's is about a third lower."

On human preference the ranking inverts again. The LMArena text leaderboard places Muse Spark 1.2 (xHigh) 5th at 1499±10 from 3,240 votes, ahead of Claude Opus 5 High in 9th at 1493±5 from 35,174 votes. As of early September, Muse Spark 1.3 had not accumulated enough votes to be listed. Read that gap carefully: a ±10 confidence interval against a ±5 one, on a tenth of the sample, is a tie dressed up as a rank.

Vals AI, on its own common harness, had Opus 5 at 67.21% against Muse Spark 1.2 at 57.05% as of 1 September; 1.3 was not yet scored.

The honest summary: on composite intelligence and human preference the two are close. On verified, reproducible agentic coding both now have a number, and Opus 5's is roughly half again as high.

Is Muse Spark 1.3 cheaper than Opus 5 in practice?

Yes, but by less than the price list implies. Three worked examples at list price. The Contributor column uses the meta/muse-spark-1.3-contributor SKU at $0.10 input / $0.20 output.

List price says 4.5× to 5.8× cheaper, and 62× to 167× cheaper on the Contributor tier. The measured figure is lower. Artificial Analysis records a weighted $1.60 per Intelligence Index v4.3.2 task on Muse Spark 1.3 against $5.86 on Opus 5 — a real-world gap of 3.7×, not 5.9×.

The discrepancy has a cause you can see in the same dataset: Muse Spark 1.3 emitted 170 million output tokens across the v4.3.2 index run against an 81 million median, which Artificial Analysis flags as "very verbose." A model that is 5.9× cheaper per output token but produces 2.1× more of them does not deliver 5.9× savings. Budget for roughly 3.7×, not 6×. In fairness to Meta, Opus 5 now carries the same "very verbose" flag at 140M tokens on the same run — verbosity is no longer a Muse-Spark-specific tax.

The Contributor tier is a different proposition entirely. At $0.10/$0.20 it is 125× cheaper on output than Opus 5, which changes what is economically sensible to automate rather than just trimming a bill. The trade is documented rather than implied — Meta's pricing page describes it as discounted pricing "in exchange for permission to use your prompts and completions to train future Meta models," against the Standard tier where prompts and completions "are not used to train Meta models."

There is a second limit that matters more than most people expect. Contributor is capped at 100 requests per minute against Standard's 3,000, and Meta applies those limits per team rather than per API key — so you cannot shard your way around it. That makes Contributor a prototyping and batch-experimentation tier, not a serving tier: the 125× discount is real, but you can only push about a thirtieth of the traffic through it. Most of the headline saving is not usable in production. For the data terms in full, see our Muse Spark 1.3 complete guide and, for the same tier on Meta's coding agent, what Muse Code's Contributor tier actually costs you. Do not put client code through it without reading those.

What can Muse Spark 1.3 see that Opus 5 cannot?

Video and images. Audio is listed but comes with a caveat Meta itself publishes, and it is the part of this comparison most likely to be reported wrongly.

Start with what is unambiguous. Anthropic's documentation states that all current Claude models support text and image input — there is no video path on Opus 5 at any price. For anything arriving as moving footage, Muse Spark 1.3 is the only one of the two that can take it, and both Muse SKUs carry the identical modality set, so the cheap tier is not stripped down here.

Audio is where the OpenRouter listing and Meta's own docs disagreed at launch — and the listing has since conceded the point. OpenRouter now shows Muse Spark 1.3's input modalities as text, image, video and file with no audio, and Artificial Analysis records speech input as unsupported on 1.3 while still listing it on 1.2. Meta's developer documentation said, verbatim:

"Audio understanding in Muse Spark 1.3 is currently not fully supported, and response quality for requests including audio content may be degraded."

Meta points audio work at Muse Spark 1.2 or its dedicated muse-voice-transcribe-1.0 model instead, and every audio example in its own guides is written against muse-spark-1.2 even though 1.3 is the default elsewhere. Meta's video-understanding page applies the same caveat to embedded soundtracks, so narrated video and screen recordings with commentary are caught by it too — the frames are fine, the audio track is not.

The advantage is real, then, but narrower than the spec sheet implies:

So a silent screen recording of a bug, a UI walkthrough or a product demo goes to Muse Spark 1.3 in one call, and cannot go to Opus 5 at all. A support call or an hour of standup audio should go to Muse Spark 1.2 or Voice Transcribe — not 1.3, and not Opus 5, which needs a separate speech-to-text vendor first.

It is an unusual regression to ship knowingly, and it reads as a signal: 1.3 is a coding and agentic point release, and audio was not what it was tuned for.

Does a 943,718-token output limit actually matter?

Less than the headline suggests, and the arithmetic has a catch. Muse Spark 1.3's ceiling is 943,718 output tokens against 128,000 for Opus 5, a 7.4× difference. But that output has to fit inside the same 1,048,576-token context window as your input. Ask for the maximum output and you have 104,858 tokens of input budget left — about a tenth of the window. It is a split of one pool, not extra capacity bolted on.

Two caveats keep this honest. Anthropic's Message Batches API raises the Opus 5 ceiling to 300,000 tokens with the output-300k-2026-03-24 beta header, so for asynchronous work the real gap is 3.1×, not 7.4×. And a single generation approaching a million tokens is rarely the right shape for a task — coherence degrades long before the limit.

Where the headroom earns its place is in avoiding truncation logic. Whole-repository migrations, bulk structured extraction, exhaustive translation passes and large synthetic-data runs all cross 128,000 output tokens routinely. On Opus 5 those need chunking and continuation handling — code you write, test and debug. On Muse Spark 1.3 they are one call. An engineering-time saving, not a capability unlock.

Which is better for coding and agentic work?

Claude Opus 5, of these two, and it is not close. Opus 5 scores 51.8% ± 3.4 on Terminal-Bench 4.0 across 330 trials — a run that consumed 6.5 billion tokens and $5,969, with an average trial duration of about 80 minutes. That is a long-horizon agentic benchmark run to completion by a third party and published with a confidence interval. It was first place when it landed in July; by October it is tenth on a much longer board, behind GPT-6 Astra at 58.2% and three Claude Fable 5.1 configurations, among others. Muse Spark 1.3 has no entry on that board, none on SWE-bench Verified, and Meta has published no launch benchmark post for 1.3 at all.

That last point is a change from 1.2. In August, Meta published a benchmark chart with 1.2 in second place behind Opus 5, plus a methodology note conceding the comparison was "not harness-identical to the leaderboard" — we took it apart in our Muse Spark 1.2 benchmarks vs Claude Opus 5 analysis. For 1.3 there is no equivalent: as of early September, Meta's AI blog index still showed "Introducing Muse Spark 1.1" from 9 July as its most recent Muse Spark post, and the developer model page carries pricing and three positioning claims — "trained for long-horizon, agentic workflows," "competitive coding performance," "native multimodal perception" — with no numbers behind them.

So there is no vendor claim to verify. Artificial Analysis is the only independent capability signal on 1.3 — an index of 48.1 on v4.3.2, with a 33.3% Terminal-Bench 4.0 pass inside it against Opus 5's 49.0%. If you need a model to drive a terminal agent unsupervised for an hour, Opus 5 has the stronger of the two numbers, and Claude Opus 5.5 — 59.6% on the same Artificial Analysis Terminal-Bench 4.0 run, at a lower list price than Opus 5 — has a considerably stronger one than either.

Where is each model genuinely worse?

Muse Spark 1.3 is worse at:

• Verified agentic coding. Absent from the official Terminal-Bench 4.0 and SWE-bench Verified boards. Artificial Analysis's own Terminal-Bench 4.0 pass scores it 33.3%, against 49.0% for Opus 5 and 59.6% for Opus 5.5.
• Latency. 47.9 s to first token on Artificial Analysis v4.3.2 — more than four times Muse Spark 1.2's 11.7 s. Whatever 1.3 gained in intelligence, it spends in thinking time before it answers.
• Verbosity. 170M output tokens per index run against an 81M median. Costs more than the price list implies and produces more text to review — though Opus 5's 140M earns the same flag.
• Audio. Documented by Meta as "not fully supported" on 1.3, with audio work routed back to 1.2 — a regression against the previous version, and now dropped from 1.3's modality list on both OpenRouter and Artificial Analysis.
• Track record. Our earlier analysis of the 1.1 launch found Meta's claimed 80.0% on Terminal-Bench 2.1 sitting above the independently verified 76.2% ± 1.2 — a small gap, but in a consistent direction across releases.

Claude Opus 5 is worse at:

• Price. 3.7× more expensive per measured task, before you consider the Contributor tier — and more expensive per token than Anthropic's own newer Opus 5.5.
• Throughput. 56.0 tokens/second against 145.9 — a 2.6× difference that users feel directly in interactive tools.
• Latency at max effort. 59.2 s to first token on Artificial Analysis v4.3.2 — still the slowest of the three configurations compared here.
• Media inputs. No audio, no video, at any tier.
• Output ceiling. 128,000 tokens synchronously, 300,000 on batch.
• Lifecycle. Anthropic lists Opus 5 under "Legacy models (still available)" and recommends starting with Opus 5.5 — cheaper at $4.00 / $20.00 and 57.6 on Artificial Analysis v4.3.2 against Opus 5's 50.8.

Which should you pick?

One question decides it: does anything unsupervised depend on the output being right?

If yes — an agent editing a repository, a production migration, a code review nobody reads afterwards — use Claude Opus 5, or Claude Opus 5.5 if you are starting fresh rather than auditing something already in production. Between these two specifically, Opus 5 has the stronger agentic-coding number by 15.7 points and the price gap is trivial next to an unsupervised mistake in a codebase. If no — classification, summarisation, extraction, drafting, anything with a human or a test suite in between — use Muse Spark 1.3. Under three index points is not worth 3.7× the bill on work that gets checked.

A practical hybrid: route the bulk stage to Muse Spark 1.3 and the verification stage to Opus 5. On the agent-loop example above that is roughly $2.35 plus a small Opus 5 check rather than $10.00 end to end, and the verified model still sits where the decision happens.

Recheck this often — and always check which index version a score came from. Artificial Analysis rebased from v4.1.1 to v4.3.2 between this article's first publication and this update, and that rebase moved more numbers on this page than any model release did: Muse Spark 1.3 reads 62 on the old index and 48.1 on the new one, with nothing about the model having changed. On the current index Muse Spark has gained 14.4 points across three releases since mid-July — 33.7, then 39.6, then 48.1 — and closed the gap to Opus 5 to 2.7 points. It is closing from one direction. Anthropic's answer was to ship Opus 5.5 at 57.6 and reopen it.

FAQ

Is Muse Spark 1.3 better than Claude Opus 5?

Not overall, but it is close. On Artificial Analysis's Intelligence Index v4.3.2 (October 2026) Muse Spark 1.3 scores 48.1 against 50.8 for Claude Opus 5. On agentic coding the gap is much wider: Artificial Analysis's own Terminal-Bench 4.0 pass puts Opus 5 at 49.0% and Muse Spark 1.3 at 33.3%, and no Muse Spark model appears on the official Terminal-Bench leaderboard at all. Muse Spark 1.3 wins on price, speed and media inputs.

Is Muse Spark 1.3 cheaper than Opus 5?

Substantially. List price is $1.25/$4.25 per million input/output tokens against $5.00/$25.00 for Opus 5 — 4× and 5.9× cheaper respectively. Measured end to end the gap narrows to about 3.7× — $1.60 against $5.86 per Intelligence Index v4.3.2 task — because Muse Spark 1.3 is verbose and emits roughly 2.1× the median number of output tokens. The Contributor tier is cheaper again at $0.10/$0.20.

Which is better for coding?

Claude Opus 5 of the two, on current evidence — though neither is the strongest option available in October 2026. Opus 5 scores 51.8% ± 3.4 on Terminal-Bench 4.0 across 330 trials with the Claude Code harness, which now places it tenth rather than first; GPT-6 Astra leads at 58.2%. Muse Spark 1.3 has no entry on the official board, and Artificial Analysis's in-house Terminal-Bench 4.0 run scores it 33.3% against Opus 5's 49.0% and Claude Opus 5.5's 59.6%.

Does Muse Spark 1.3 support audio?

No longer, even nominally. Audio was briefly a listed input type, but Meta's own docs say "audio understanding in Muse Spark 1.3 is currently not fully supported, and response quality for requests including audio content may be degraded," and point you to Muse Spark 1.2 or muse-voice-transcribe-1.0 instead. OpenRouter and Artificial Analysis have both since removed audio from 1.3's listed modalities, while still showing speech input on 1.2. The caveat also covers soundtracks embedded in video. Video and image input, by contrast, work fully — and Claude Opus 5 has no video path at all.

What is the Muse Spark Contributor tier?

A second SKU at $0.10 input and $0.20 output per million tokens — about 125× cheaper on output than Opus 5 — with identical context, output limit and modalities. Meta grants the discount "in exchange for permission to use your prompts and completions to train future Meta models." It is also rate-limited to 100 requests per minute against Standard's 3,000, applied per team, which makes it a prototyping tier rather than a serving tier.

Should I switch from Opus 5 to Muse Spark 1.3?

Switch the workloads where output gets checked by a human or a test suite — bulk classification, summarisation, extraction, drafting. Keep Opus 5 for anything unsupervised that touches a codebase, since it is the only one of the two with a verified long-horizon agentic result. A hybrid split, bulk on Muse Spark and verification on Opus 5, captures most of the saving.

How big is the context and output difference?

Context windows are nearly identical: 1,048,576 tokens for Muse Spark 1.3 against 1,000,000 for Opus 5. Max output differs sharply — 943,718 versus 128,000 synchronously, though Anthropic's Message Batches API raises Opus 5 to 300,000 with a beta header. Note that Muse Spark's output shares the same window, so a maximum-length generation leaves only about 104,858 tokens for input.

Where does Muse Spark 1.3 rank against other models?

Artificial Analysis places the max-effort variant #23 of the 224 models in its comparison class, at an index of 48.1 on v4.3.2 against a class median of 26. On the official Terminal-Bench 4.0 board there is no Muse Spark entry at all. For the wider Muse family and how the versions differ, see our Muse Spark complete guide.

Why do other sites quote Muse Spark 1.3 at 62 on Artificial Analysis?

Because they are citing Intelligence Index v4.1.1, which Artificial Analysis has retired. v4.3.2 rebased the scale and changed the constituent evaluations — it now runs AA-Briefcase v1.1, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, Humanity's Last Exam, GDP.pdf, CritPt, AA-Omniscience and AA-LCR v1.1. There is no conversion factor between the versions: 62 was correct on v4.1.1 and 48.1 is correct on v4.3.2, for the same model on the same weights. Any score quoted without its index version is unusable.

Is Claude Opus 5 still the Anthropic model to use?

No. Anthropic lists Claude Opus 5 under "Legacy models (still available)" and advises starting with Claude Opus 5.5, which is cheaper at $4.00 / $20.00 per million tokens against Opus 5's $5.00 / $25.00, scores 57.6 on Artificial Analysis's Intelligence Index v4.3.2 against 50.8, and reaches 59.6% on Artificial Analysis's Terminal-Bench 4.0 pass against 49.0%. Opus 5 remains available and callable, so this comparison still matters if you are already running it — but it is not what you would choose today.

If you are still on the previous generation, our Muse Spark 1.2 benchmarks comparison covers the upgrade case, and the Claude Opus 5 launch guide has the full Anthropic-side detail.

Updated 5 October 2026. Artificial Analysis scores re-pulled that day against Intelligence Index v4.3.2 (the figures in the first version of this article were v4.1.1 and are not comparable); Terminal-Bench 4.0 ranks, OpenRouter model listings and Anthropic's model lineup re-checked the same day. The Meta developer-docs quotes date from 3 September 2026 — Meta has since put those pages behind a sign-in wall, so they could not be re-verified, and they are reported as of that date. LMArena and Vals AI figures are also September snapshots. Benchmark leaderboards move and index versions get rebased; check both before committing a migration.