Muse Wonder vs Claude Opus 5.5
Identical $4/$20 API list prices, yet $4.52 vs $5.98 per task — how do Meta's Muse Wonder and Muse Wonder Full stack up against Claude Opus 5.5?
The short answer
| At a glance | Muse Wonder | Claude Opus 5.5 |
|---|---|---|
| List priceinput / output per 1M tokens | $4 / $20 | $4 / $20 |
| Benchmark task costArtificial Analysis, per Intelligence-Index task | $4.52 (better) | $5.98 |
| Intelligence IndexAA v4.3.2, higher is better | 52 | 58 (better) |
Meta's Muse Wonder (muse-wonder-1.0, launched October 12, 2026) matches Claude Opus 5.5's exact standard list price: $4.00 per 1M input tokens, $20.00 per 1M output tokens, and $0.20 per 1M cached-input tokens across a flat 1M context window. In real workloads, however, three structural differences separate them: (1) Per-task verbosity & latency — on Artificial Analysis's independent Intelligence Index v4.3.2, Muse Wonder emits 54k output tokens/task (29k reasoning) vs Opus 5.5's 119k (84k reasoning), cutting average benchmark task cost by 24% ($4.52 vs $5.98) and Time to First Answer Token by 12x (60.82s vs 748.61s); (2) API vs subscription tiering — Meta's discounted Contributor mode is NOT available for muse-wonder-1.0 on the Model API, and the higher-scoring Muse Wonder Full (AA Index 56) is locked to Muse Code Power/Maximum Usage subscriptions with no pay-as-you-go API; and (3) Task-shaped quality — Meta's announcement table puts Muse Wonder ahead on DeepSWE v1.1 (78.8% vs 74.2%), SWE-Marathon (53.1% vs 46.3%), MemoryBench (85.6% vs 81.5%), and Harvey LAB-AA (10.0% vs 4.2%), while independent AA v4.3.2 gives Opus 5.5 a 6-point overall lead (58 vs 52) across Terminal-Bench 4.0 (60% vs 52%), SciCode (67% vs 60%), HLE (61% vs 50%), and GDPval-AA (1866 vs 1770), with Muse Wonder winning GDP.pdf (31% vs 26%).
Choose Muse Wonder
Lower per-task spend at the $4/$20 tier ($4.52 vs $5.98 per AA task), 12x faster time to first answer token (60.82s vs 748.61s), zero cache-write fees on prefix caching, PDF reasoning (GDP.pdf 31% vs 26%), or repository/memory workflows where Meta's DeepSWE (78.8%), SWE-Marathon (53.1%), and MemoryBench (85.6%) leads hold up on your repo.
Choose Claude Opus 5.5
Highest independent overall quality (AA Intelligence Index 58 vs 52, GDPval-AA 1866 vs 1770, AA-Briefcase 1807 vs 1659, Terminal-Bench 4.0 60% vs 52%, SciCode 67% vs 60%), 50%-off Batch API ($2/$10) or Fast mode (2x), or when you need API access to a 56+ Index model (since Muse Wonder Full has no API).
Either — decide by evidence you generate
OpenAI-SDK or multi-provider routing at flat 1M context: both charge $4/$20 standard and $0.20 cached reads with no long-context cliff.
Key facts
- Identical standard-tier list prices: $4.00 / 1M input, $20.00 / 1M output, and $0.20 / 1M cached-input tokens across a flat 1,000,000-token context window.
- Measured per-task cost (AA Intelligence Index v4.3.2): Muse Wonder (Max) averages $4.52/task ($5,733 to run the full Index, 54k output tokens/task) vs Claude Opus 5.5's $5.98/task ($8,708 total, 119k output tokens/task) — 24% cheaper per task at identical token prices.
- Two critical Meta API caveats: (1) Contributor mode (Meta's discounted data-sharing tier available on muse-spark-1.3 at $0.10/$0.20) is explicitly NOT available for muse-wonder-1.0 on Model API; (2) Muse Wonder Full (AA Index 56, DeepSWE 79.0%) has NO pay-as-you-go API and is exclusive to Muse Code Power/Maximum Usage subscriptions.
- Caching & service-tier gap: Meta Model API applies a 95% discount ($0.20/MTok) on prefix cache hits with no published cache-write fee, but offers only Standard on-demand serving (3,000 RPM / 4M TPM). Claude Opus 5.5 charges $5.00/MTok for 5-minute cache writes ($8.00 for 1-hour), but offers a 50%-off Batch API ($2/$10) and 2x Fast mode.
- Independent vs vendor benchmarks: On AA v4.3.2 (same harness), Opus 5.5 leads overall Index (58 vs 52), GDPval-AA (1866 vs 1770), AA-Briefcase (1807 vs 1659), Terminal-Bench 4.0 (60% vs 52%), SciCode (67% vs 60%), and HLE (61% vs 50%), while Muse Wonder leads GDP.pdf (31% vs 26%) and Time to First Answer Token (60.82s vs 748.61s). In Meta's announcement table, Muse Wonder leads DeepSWE v1.1 (78.8% vs 74.2%), SWE-Marathon (53.1% vs 46.3%), MemoryBench (85.6% vs 81.5%), and Harvey LAB-AA (10.0% vs 4.2%).
Official pricing, every tier
USD per 1M tokens from vendor pricing pages. 'Not published' means exactly that — we don't extrapolate.
| Service tier | Muse Wonder | Claude Opus 5.5 |
|---|---|---|
| Standarddefault on-demand serving | in $4 · cached $0.20 · out $20 implicit cache only (no write fee) | in $4 · cached $0.20 · out $20 cache write $5 |
| Batchasync jobs, 50% off | Not published | in $2 · cached $0.10 · out $10 cache write $2.5 · derived |
| Fastpriority speed, 2x (ex-Priority) | Not published | in $8 · cached $0.40 · out $40 cache write $10 · derived |
| Long-context band | Muse Wonder | Claude Opus 5.5 |
|---|---|---|
| Long-context ratesapplies per request above the model threshold | Flat pricing same rates at any input size (up to context window) | Flat pricing same rates at any input size (up to context window) |
USD per 1M tokens. Every input token is billed in exactly one bucket — uncached, cached read, or cache write; a write replaces the input rate for those tokens (official rule), so write rates are per tier as listed. Anthropic rows bill the 5-minute cache-write TTL; the 1-hour TTL ($8/MTok standard) is also published, and cache multipliers stack with the Batch discount and Fast-mode pricing. · Anthropic pricing (verified 2026-10-09) · Meta Muse Wonder (muse-wonder-1.0) is billed in Standard mode only ($4/$20, $0.20 cached reads, flat 1M context); Contributor mode is not available for Muse Wonder on Model API, and Muse Wonder Full is exclusive to Muse Code subscriptions. · Meta Model API pricing (verified 2026-10-13)
Tier ≠ quality change
Run this comparison on your own numbers
Preloaded with a workload typical for this comparison. Change anything — results recompute locally.
Token-based workload estimate only; it does not measure quality or cost per successful task.
| Model | Per request | Monthly |
|---|---|---|
| Muse Wonder cheapest | $0.218 | $2,180.00 |
| Claude Opus 5.5 | $0.218 | $2,180.00 |
Monthly spread between the cheapest and priciest selected model: $0.00.
Breakdown: Muse Wonder · standard tier · standard context
- Fresh input (uncached): 30,000 tok × $4/M$0.12
- Cached input (reads): 90,000 tok × $0.2/M$0.018
- Cache writes: 0 tok × $0/M$0.00
- Output (incl. reasoning tokens): 4,000 tok × $20/M$0.08
- Per request$0.218
Rates: official pricing · verified 2026-10-13
- Per-token arithmetic only: it does not model quality differences or cross-vendor output-token-count differences for the same task.
- Reasoning tokens are billed as output tokens at the chosen model's output rate.
- Long-context band is decided per request by total input tokens vs the provider threshold, not by monthly volume.
- Cache accounting uses the official mutually-exclusive buckets: each input token is billed once, as uncached input, cached read, OR cache write — writes replace the input rate for those tokens, never add to it.
- Claude rows bill the 5-minute-TTL cache write; the published 1-hour TTL ($8/MTok standard) is not modeled (no TTL input).
- Rows flagged 'derived' are computed from a stated official rule (e.g. Anthropic batch -50%), not read verbatim from a price table.
- Missing official prices render as Unsupported — we never apply a guessed multiplier (e.g., GPT-6 Sol has no Ultrafast tier).
What the evidence actually says
Independent and vendor-run evals, each with publisher, runner, and harness. Numbers from different runs are never merged.
| Benchmark | Score | Effort | Harness | Evidence |
|---|---|---|---|---|
| AutomationBenchunspecified⚠ run comparability unknown | ||||
| claude-opus-5-5 | 40 % | unspecified | Zapier (Opus 5.5 measured during Zapier's early access), published by Anthropic | Third-party |
| gpt-6-astra | 41.4 % | unspecified | Zapier public leaderboard, cited by Anthropic | Third-party |
| Terminal-Bench-Science0.1⚠ run comparability unknown | ||||
| gpt-6-astra | 68.1 % | unspecified | OpenAI announcement evals | Vendor-reported |
| gpt-6-astra | 64.6 % | unspecified | OpenAI-reported figure, cited by Anthropic (footnote 3) | Vendor-reported |
| claude-opus-5-5 | 58.7 % | max (adaptive thinking) | Anthropic announcement evals | Vendor-reported |
| Terminal-Bench-Science cost per task0.1⚠ run comparability unknown | ||||
| claude-opus-5-5 | 23.21 usd-per-task | unspecified | OpenAI announcement cost table (who measured the competitor's cost is not stated) | Vendor-reported |
| gpt-6-astra | 23.8 usd-per-task | unspecified | OpenAI announcement evals | Vendor-reported |
| Terminal-Bench4.0⚠ run comparability unknown | ||||
| claude-opus-5-5 | 66.4 % | xhigh (adaptive thinking) | Anthropic announcement evals | Vendor-reported |
| gpt-6-astra | 57.9 % | high (as reported by OpenAI) | OpenAI-reported figure, cited by Anthropic (footnote 1) | Vendor-reported |
| claude-opus-5-5 | 60 % | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 52 % | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| FrontierCodev1.1 (Main)⚠ run comparability unknown | ||||
| claude-opus-5-5 | 54.4 % | max (adaptive thinking) | Anthropic announcement evals | Vendor-reported |
| claude-opus-5-5 | 54.6 % | medium (default) | Anthropic announcement evals | Vendor-reported |
| gpt-6-astra | 53.3 % | unspecified | Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run) | Vendor-reported |
| CursorBench4.0 | ||||
| claude-opus-5-5 | 57.8 % | max (adaptive thinking) | Anthropic announcement evals | Vendor-reported |
| GDPval-AAv2.1⚠ run comparability unknown | ||||
| claude-opus-5-5 | 1846 elo | unspecified | Anthropic announcement evals | Vendor-reported |
| gpt-6-astra | 1542 elo | unspecified | Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run) | Vendor-reported |
| claude-opus-5-5 | 1866 elo | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 1770 elo | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| Humanity's Last Exam (with tools)unspecified⚠ run comparability unknown | ||||
| claude-opus-5-5 | 67.7 % | unspecified | Anthropic announcement evals | Vendor-reported |
| gpt-6-astra | 57.2 % | unspecified | Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run) | Vendor-reported |
| OSWorld2.1 (partial) | ||||
| claude-opus-5-5 | 81.8 % | unspecified | Anthropic announcement evals | Vendor-reported |
| AA Intelligence Indexv4.3.2 | ||||
| claude-opus-5-5 | 58 index-points | config 'Opus 5.5 (Max, Default Fallback)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 52 index-points | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| SciCodeunspecified | ||||
| claude-opus-5-5 | 67 % | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 60 % | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| AA-Briefcasev1.1 | ||||
| claude-opus-5-5 | 1807 elo | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 1659 elo | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| Humanity's Last Examunspecified | ||||
| claude-opus-5-5 | 61 % | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 50 % | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| AutomationBench-AAunspecified | ||||
| claude-opus-5-5 | 70 % | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 68 % | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| AA cost per Intelligence-Index taskv4.3.2 | ||||
| claude-opus-5-5 | 5.98 usd-per-task | — | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models) | Third-party |
| muse-wonder-1.0 | 4.52 usd-per-task | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models) | Third-party |
| Vals Index (DeepMind table)Oct 2026 | ||||
| claude-opus-5-5 | 67 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| gpt-6-astra | 63.1 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| GraphWalks (256K–1M long context)BFS F1 | ||||
| claude-opus-5-5 | 66.8 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| gpt-6-astra | 71.8 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| LABBench 2v2 | ||||
| claude-opus-5-5 | 73.1 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| gpt-6-astra | 85.4 % | — | Google DeepMind Model Card evaluation table (competitor column) | Vendor-reported |
| GDP.pdfunspecified | ||||
| muse-wonder-1.0 | 31 % | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness) | Third-party |
| claude-opus-5-5 | 26 % | config 'Opus 5.5 (Max, Default Fallback)' per AA | Artificial Analysis Intelligence Index v4.3.2 (independent, same harness) | Third-party |
| AA Time to First Answer Tokenv4.3.2 | ||||
| muse-wonder-1.0 | 60.82 seconds | config 'Muse Wonder (Max)' per AA | Artificial Analysis Intelligence Index v4.3.2 latency test | Third-party |
| claude-opus-5-5 | 748.61 seconds | config 'Opus 5.5 (Max, Default Fallback)' per AA | Artificial Analysis Intelligence Index v4.3.2 latency test | Third-party |
| DeepSWEv1.1 | ||||
| muse-wonder-1.0 | 78.8 % | max | Meta Muse Wonder announcement evaluation table | Vendor-reported |
| claude-opus-5-5 | 74.2 % | max | Meta Muse Wonder announcement evaluation table (competitor column) | Vendor-reported |
| SWE-Marathonunspecified | ||||
| muse-wonder-1.0 | 53.1 % | max | Meta Muse Wonder announcement evaluation table | Vendor-reported |
| claude-opus-5-5 | 46.3 % | max | Meta Muse Wonder announcement evaluation table (competitor column) | Vendor-reported |
| MemoryBenchunspecified | ||||
| muse-wonder-1.0 | 85.6 % | max | Meta Muse Wonder announcement evaluation table | Vendor-reported |
| claude-opus-5-5 | 81.5 % | max | Meta Muse Wonder announcement evaluation table (competitor column) | Vendor-reported |
| Harvey LAB-AAunspecified | ||||
| muse-wonder-1.0 | 10 % | max | Meta Muse Wonder announcement evaluation table | Vendor-reported |
| claude-opus-5-5 | 4.2 % | max | Meta Muse Wonder announcement evaluation table (competitor column) | Vendor-reported |
Same benchmark name, multiple rows — run comparability unknown
AutomationBench: claude-opus-5-5 = 40 (Zapier (Opus 5.5 measured during Zapier's early access), published by Anthropic); gpt-6-astra = 41.4 (Zapier public leaderboard, cited by Anthropic). Zapier-run. OpenAI's announcement reports a +2.2 pp delta for GPT-6.1 Sol over Opus 5.5 with an unstated underlying runner — the two publications' run relationship is unknown; do not combine.
Terminal-Bench-Science: gpt-6-astra = 68.1 (OpenAI announcement evals); gpt-6-astra = 64.6 (OpenAI-reported figure, cited by Anthropic (footnote 3)); claude-opus-5-5 = 58.7 (Anthropic announcement evals). Anthropic's announcement also carries an Astra TBS figure of 64.6%, footnoted there as 'as reported by OpenAI' — BOTH figures are OpenAI-published, yet they differ (68.1 vs 64.6). Which runs/versions/dates produced each figure is not documented in either announcement; we store both and merge nothing.
Terminal-Bench-Science cost per task: claude-opus-5-5 = 23.21 (OpenAI announcement cost table (who measured the competitor's cost is not stated)); gpt-6-astra = 23.8 (OpenAI announcement evals). Whether these rows share a run is not documented in the sources; treat them as not comparable.
Terminal-Bench: claude-opus-5-5 = 66.4 (Anthropic announcement evals); gpt-6-astra = 57.9 (OpenAI-reported figure, cited by Anthropic (footnote 1)); claude-opus-5-5 = 60 (Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models)); muse-wonder-1.0 = 52 (Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models)). Whether these rows share a run is not documented in the sources; treat them as not comparable.
FrontierCode: claude-opus-5-5 = 54.4 (Anthropic announcement evals); claude-opus-5-5 = 54.6 (Anthropic announcement evals); gpt-6-astra = 53.3 (Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run)). Whether these rows share a run is not documented in the sources; treat them as not comparable.
GDPval-AA: claude-opus-5-5 = 1846 (Anthropic announcement evals); gpt-6-astra = 1542 (Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run)); claude-opus-5-5 = 1866 (Artificial Analysis Intelligence Index v4.3.2 (independent, same harness for all models)); muse-wonder-1.0 = 1770 (Artificial Analysis Intelligence Index v4.3.2 (independent, same harness across models)). Whether these rows share a run is not documented in the sources; treat them as not comparable.
Humanity's Last Exam (with tools): claude-opus-5-5 = 67.7 (Anthropic announcement evals); gpt-6-astra = 57.2 (Anthropic announcement evals (competitor row; no runner footnote — assumed Anthropic-run)). Whether these rows share a run is not documented in the sources; treat them as not comparable.
We never average or merge these. See methodology.Sources: Anthropic — Introducing Claude Opus 5.5 · OpenAI — Introducing GPT-6.1 Sol (DevDay 2026) · Artificial Analysis (independent) — Independent same-harness comparison: GPT-6.1 Sol vs Claude Opus 5.5 (Intelligence Index v4.3.2) · Google DeepMind — Gemini 4 Argon Model Card (October 2026 evaluation table) · Artificial Analysis (independent) — Independent same-harness comparison: Muse Wonder (Max) vs Claude Opus 5.5 (Intelligence Index v4.3.2) · Meta — Put your ideas to work with Muse Wonder and Muse Wonder Full (official release blog & benchmark table)
Which model for which workload
Confidence labels: Measured = same-harness evidence · Inferred = reasoned from documented evidence · Insufficient evidence = no comparable public data.
| Workload | Our pick | Confidence | Why |
|---|---|---|---|
| Daily coding tasks | Muse Wonder | Measured | At identical $4/$20 rates, Muse Wonder emits less than half as many reasoning tokens per task ($4.52 vs $5.98 on AA) and answers 12x faster (60.82s vs 748.61s TTFAT), while scoring 78.8% on DeepSWE v1.1 in Meta's table. |
| Complex repository maintenance | Muse Wonder | Measured | Meta's evaluation table shows Muse Wonder ahead on long-horizon SWE-Marathon (53.1% vs 46.3%), DeepSWE v1.1 (78.8% vs 74.2%), and MemoryBench (85.6% vs 81.5%) with zero cache-write fees across 1M context. Vendor-reported table; Opus 5.5 leads independent AA SciCode (67% vs 60%) and Terminal-Bench 4.0 (60% vs 52%). |
| Agent workflows | Claude Opus 5.5 | Measured | Opus 5.5 leads independent same-harness AA Terminal-Bench 4.0 (60% vs 52%), GDPval-AA (1866 vs 1770), AA-Briefcase (1807 vs 1659), and AutomationBench-AA (70% vs 68%). |
| Long-context tasks | Muse Wonder | Measured | Both offer flat $4/$20 pricing to 1M tokens, but Muse Wonder leads AA GDP.pdf (31% vs 26%) and Meta's MemoryBench (85.6% vs 81.5%) and avoids Opus 5.5's $5.00/MTok cache-write charge on large prefixes. |
| Quality-first | Claude Opus 5.5 | Measured | Leads AA Intelligence Index v4.3.2 by 6 points (58 vs 52 for API Muse Wonder, and 58 vs 56 for subscription-only Muse Wonder Full), including a +11 pp lead on Humanity's Last Exam (61% vs 50%). |
| Budget-sensitive / high volume | Muse Wonder | Measured | On Standard on-demand API calls, Muse Wonder's lower reasoning verbosity saves 24% per task ($4.52 vs $5.98 on AA) and eliminates cache-write fees. |
| Offline batch jobs | Claude Opus 5.5 | Price-derived | Anthropic publishes a 50%-off Batch API ($2.00/$10.00, $0.10 cached read); Meta Model API currently publishes only Standard on-demand pricing ($4.00/$20.00) for muse-wonder-1.0. |
| Latency-sensitive production | Muse Wonder | Measured | On AA v4.3.2, Muse Wonder (Max) delivers its first answer token in 60.82s vs 748.61s on Claude Opus 5.5 (Max) because it emits 29k reasoning tokens per task instead of 84k. |
Frequently asked questions
What is the API price of Meta Muse Wonder (muse-wonder-1.0)?
On the Meta Model API (dev.meta.ai/docs/pricing-rate-limits), muse-wonder-1.0 is priced at $4.00 per 1M input tokens, $20.00 per 1M output tokens, and $0.20 per 1M cached-input tokens (a 95% discount on prefix cache hits, with no separate cache-write fee) across its 1,000,000-token context window. That is the exact same headline list price as Claude Opus 5.5.
Can I use Meta's cheaper Contributor mode or Muse Wonder Full via the API?
No to both. Per Meta's official pricing docs and release blog (2026-10-12): (1) Contributor mode (the discounted data-sharing tier that drops muse-spark-1.3 from $1.25/$4.25 to $0.10/$0.20) is explicitly NOT available for muse-wonder-1.0 on the Model API — it is only available inside Muse Code subscriptions; and (2) Muse Wonder Full (which scores 56 on the AA Intelligence Index vs 52 for muse-wonder-1.0) has no pay-as-you-go API and is exclusive to Muse Code Power Usage and Maximum Usage plans.
Why does Muse Wonder cost $4.52 per task vs Claude Opus 5.5's $5.98 when both charge $4/$20?
Reasoning token verbosity. On Artificial Analysis's Intelligence Index v4.3.2, both models are billed at $4.00 input / $20.00 output / $0.20 cached input, but Muse Wonder (Max) emits an average of 54k output tokens per task (29k reasoning tokens, 134M total output tokens across the suite) compared with 119k output tokens per task on Claude Opus 5.5 (84k reasoning tokens, 260M total). Fewer thinking tokens at $20/MTok translates to a 24% lower bill per task ($4.52 vs $5.98) and 12x faster time to first answer token (60.82s vs 748.61s).
How does Muse Wonder compare to Claude Opus 5.5 on coding and agentic benchmarks?
It depends on the harness. In Meta's announcement evaluation table (2026-10-12), Muse Wonder (Max) beats Claude Opus 5.5 (Max) on DeepSWE v1.1 (78.8% vs 74.2%), SWE-Marathon (53.1% vs 46.3%), MemoryBench (85.6% vs 81.5%), and Harvey LAB-AA (10.0% vs 4.2%). In Artificial Analysis's independent same-harness Intelligence Index v4.3.2, Claude Opus 5.5 leads overall (58 vs 52), Terminal-Bench 4.0 (60% vs 52%), SciCode (67% vs 60%), GDPval-AA (1866 vs 1770), and AA-Briefcase (1807 vs 1659), while Muse Wonder wins GDP.pdf (31% vs 26%).
Related
Sources & freshness
- Meta — Pricing and rate limits - Meta Model API (muse-wonder-1.0 standard pricing, caching, contributor mode)https://dev.meta.ai/docs/pricing-rate-limits · accessed 2026-10-13Muse Wonder (muse-wonder-1.0) Standard mode: $4.00 / 1M input tokens, $20.00 / 1M output tokens, $0.20 / 1M cached input tokens (95% discount on prefix cache hits; no explicit cache-write fee published; flat across 1,000,000-token context window). Contributor mode (discounted data-sharing mode available on muse-spark-1.3 at $0.10/$0.20) is explicitly NOT available for muse-wonder-1.0 on Model API (available only in Muse Code subscriptions). Rate limits: 3,000 RPM, 4,000,000 TPM per team. Web search grounding: $2.50 / 1,000 queries.
- Anthropic — Claude models overview (model IDs, context window, max output, retirement)https://platform.claude.com/docs/en/docs/about-claude/models/overview · accessed 2026-10-09claude-opus-5-5: 1M context, 128K max output, training cutoff Jun 2026, retirement not sooner than 2027-09-22, cache reads '5% of input price', batch 50% off.
- Anthropic — Claude pricing (API pricing table incl. cache write/read, fast mode, batch)https://claude.com/pricing · accessed 2026-10-09Opus 5.5: $4/$20 per MTok, cache read $0.20, cache write (5m TTL) $5, batch 'save 50%', fast mode 2x on input/output.
- Anthropic — Claude pricing docs (cache TTL prices, fast/batch stacking rules)https://platform.claude.com/docs/en/about-claude/pricing · accessed 2026-10-09Opus 5.5 1h cache write $8/MTok; caching multipliers apply on top of fast-mode pricing and stack with the Batch discount; fast mode not available with Batch.
- Anthropic — Introducing Claude Opus 5.5https://www.anthropic.com/claude-opus-5-5 · published 2026-09-22 · accessed 2026-10-09Self-reported evals (Anthropic harness, adaptive thinking, max effort unless noted). Includes competitor scores incl. GPT-6 Astra as measured by Anthropic.
- OpenAI — Introducing GPT-6.1 Sol (DevDay 2026)https://openai.com/index/introducing-gpt-6-1-sol · published 2026-09-29 · accessed 2026-10-09Self-reported evals (OpenAI harness). Includes per-task cost quotes (Terminal-Bench-Science). Announces GPT-6.1 Sol Ultrafast 'coming soon' in Codex. Availability: ChatGPT Work + Codex, not yet in Chat.
- Artificial Analysis (independent) — Independent same-harness comparison: GPT-6.1 Sol vs Claude Opus 5.5 (Intelligence Index v4.3.2)https://artificialanalysis.ai/models/comparisons/gpt-6-1-sol-vs-claude-opus-5-5 · accessed 2026-10-10Independent same-harness evaluation of both models. Cost-per-task = weighted average per Intelligence Index task; reflects measured token verbosity. Effort configs as labeled by AA ('Sol (Max)' vs 'Opus 5.5 (Max, Default Fallback)').
- Google DeepMind — Gemini 4 Argon Model Card (October 2026 evaluation table)https://deepmind.google/models/model-cards/gemini-4-argon/ · published 2026-10-13 · accessed 2026-10-13Official evaluation table comparing Gemini 4 Argon, GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across Vals Index (68.9% vs 67.0% Opus), DeepSWE v1.1 (77.9% vs 74.2% Opus), Terminal-Bench 4.0 (57.4% vs 66.4% Opus), GraphWalks 256k-1M (84.2% vs 66.8% Opus), LABBench 2 (88.8% vs 73.1% Opus), and RiemannBench (76.0% vs 69.6% Opus). Knowledge cutoff March 2026.
- Artificial Analysis (independent) — Independent same-harness comparison: Muse Wonder (Max) vs Claude Opus 5.5 (Intelligence Index v4.3.2)https://artificialanalysis.ai/models/comparisons/muse-wonder-vs-claude-opus-5-5 · published 2026-10-12 · accessed 2026-10-13Independent same-harness Intelligence Index v4.3.2 evaluation of Muse Wonder (Max): Index 52 (vs 58 Opus 5.5, 53 Gemini 4 Argon, 52 GPT-6.1 Sol; subscription-only Muse Wonder Full scores 56), cost per Intelligence-Index task $4.52 (54k output tokens/task, 134M output tokens to run Index, $5,733 total) vs $5.98 Opus 5.5 (119k output tokens/task, 260M total), output speed 85 tok/s, time to first answer token 60.82s vs 748.61s Opus 5.5.
- Meta — Put your ideas to work with Muse Wonder and Muse Wonder Full (official release blog & benchmark table)https://dev.meta.ai/resources/blog/build-with-muse-wonder · published 2026-10-12 · accessed 2026-10-13Announces Muse Wonder (muse-wonder-1.0 on Meta Model API and Muse Code) and Muse Wonder Full (exclusively in Muse Code Power/Maximum Usage subscriptions, no pay-as-you-go API). Official evaluation table compares Muse Wonder Full (max), Muse Wonder (max), Claude Fable 5.1 (max), GPT-6 Astra (max), and Claude Opus 5.5 (max) across DeepSWE 1.1 (78.8% Wonder vs 74.2% Opus), SWE-Marathon (53.1% vs 46.3%), MemoryBench (85.6% vs 81.5%), Harvey LAB-AA (10.0% vs 4.2%), Terminal-Bench 4.0-AA (52.0% vs 59.6%), OSWorld V2.1 partial (72.6% vs 81.8%), and GDP.pdf (30.8% vs 30.6%).
Pricing data last verified 2026-10-13. OpenAI ships roughly weekly — re-verify before making spend commitments. See the methodology page for update cadence and stale-data flags.