Model comparison
Muse Spark 1.3 vs Qwen 3.8 Max
Meta against Qwen, compared on context, price, and verified benchmark results.
Catalog record checked September 5, 2026Individual provider fields may changeEvidence confidence: High — see receipts below
Meta
Muse Spark 1.3
Frontier
Qwen
Qwen 3.8 Max
Frontier · Open weights
| Specification | Muse Spark 1.3 | Qwen 3.8 Max |
|---|---|---|
| Provider | Meta | Qwen |
| Tier | Frontier | Frontier |
| Context window | Winner: 1M | 991K |
| Max output | Winner: 131K | 131K |
| Input / 1M tokens | Winner: $1.25 | $2 |
| Output / 1M tokens | Winner: $4.25 | $6 |
| Weights | Closed | Open |
| Parameters | Meta frontier model | 2.4T total / 95B active (MoE) |
| Reasoning levels | low, medium, high | low, high, max |
| Modalities | text, image | text, image, video |
| API model id | muse-spark-1.3 | qwen3.8-max |
| Released | September 2, 2026 | August 3, 2026 |
| DeepSWE 1.1 (2026-09-02) | 75.4 | Not verifiedUnverified |
| Terminal-Bench 2.1 (2026-09-03) | Winner: 88.8 | 86.6 |
| Artificial Analysis Intelligence Index (2026-08-14) | Not verifiedUnverified | 58 |
| GPQA Diamond (2026-08-03) | Not verifiedUnverified | 92.6 |
| Humanity's Last Exam (2026-08-14) | Not verifiedUnverified | 56.2 |
| SWE-bench Verified (2026-09-01) | Not verifiedUnverified | 85.6 |
Prices are USD per million tokens at standard rates, excluding batch and caching discounts. Bold indicates the better figure where one is objectively better. Values we could not confirm from the provider are shown as “Not verified” rather than estimated.
Benchmark receipts
- 2026-09-03: Meta launch table (transcribed by explainx.ai) (Vendor self-report; absent from the official tbench.ai leaderboard)Agentic terminal work: multi-step tasks executed in a sandboxed shell environment. Comparability: comparable with caveat — Scores across Terminal-Bench major versions (v1 vs v2) are NOT comparable; harness configuration (container, time limits) also shifts results.
- 2026-09-02: Meta launch table (Thinking mode: max — self-reported; not broadly available and NOT on the official Datacurve leaderboard)Agentic software engineering: pass@1 on real repo tasks with reported cost per completed task. Comparability: directly comparable
- 2026-09-01: vals.ai SWE-bench Verified leaderboard (independent, mini-swe-agent bash-only); not published by Qwen (official used SWE-bench Pro 67.7)Real GitHub issue resolution: does the model's patch pass the hidden tests. Comparability: comparable with caveat — Post-audit vendor claims and pre-audit scores sit on different task trust levels; scaffolding (agent harness, compute budget) also dominates results. Never aggregate across scaffolds.
- 2026-08-14: Z.ai GLM-5.3 blog (independent Z.ai-run, with tools; Qwen official no-tools 43.6 — both preserved)Frontier-knowledge ceiling: extremely hard multi-domain questions written to be near-impossible without deep expertise. Comparability: comparable with caveat — Subset/tool configurations (text-only vs with-tools) differ between vendors and materially change scores; must match configuration to compare.
- 2026-08-14: Artificial AnalysisComposite index blending reasoning, knowledge, and coding evals into one 0–100 score. Comparability: directly comparable — AA occasionally rebaselines the index scale between snapshots — a score captured on one date is only comparable to same-snapshot scores (check measuredAt).
- 2026-08-03: Qwen official blog (vendor-run table)Agentic terminal work: multi-step tasks executed in a sandboxed shell environment. Comparability: comparable with caveat — Scores across Terminal-Bench major versions (v1 vs v2) are NOT comparable; harness configuration (container, time limits) also shifts results.
Pricing tiers: Muse Spark 1.3: $1.25/$4.25 per MTok on Meta's Model API (cached input $0.15); a $0.10/$0.20 'contributor' tier applies when your data is used for training. Third-party reports put list at $1.50/$7.50 from 2027-01-01 — confirm before long-term commits. · Qwen 3.8 Max: $2/$6 per MTok (Alibaba Cloud Model Studio). Weights announced open, verify they have dropped.
Muse Spark 1.3
Muse Spark 1.3 is Meta's September 2026 frontier refresh — a self-reported DeepSWE 1.1 field leader at $1.25/$4.25, with a 1M-token context aimed at autonomous agent workflows.
Best for
- Long-horizon coding agents
- Frontier quality below frontier pricing
- Meta ecosystem
Watch out
The headline 75.4% DeepSWE score comes from the 'max' thinking mode, which is not broadly available yet and is pending independent verification. The 131,072 output cap is per third-party API docs — Meta's own spec page does not publish it. Closed weights, unlike Muse Spark 1.1.
Qwen 3.8 Max
Qwen 3.8 Max is Alibaba's flagship — 2.4T MoE with 1M-class context, near frontier on the Intelligence Index.
Best for
- Open-weight frontier work
- Long-context
- Multimodal
Watch out
Open weights dropped 2026-08-12 under a custom (non-Apache) licence with vision and 1M-context stripped from the open checkpoint — the open checkpoint is not the full API model. Verify the licence before commercial use.
When the cheaper one wins
Muse Spark 1.3 is cheaper on output at $4.25 per million tokens against $6 for Qwen 3.8 Max — about 1.4×. Use the cheaper tier for classification, extraction, summarisation, and any task where the expensive model’s extra score does not change the accepted output. The expensive one only pays if your hardest task actually fails on the cheap tier. These are standard-tier API rates, excluding batch and cache discounts.
Run the model pickerEvidence confidence: High
How strong and complete the evidence behind this comparison is — not a prediction of which model is better.
- Pricing verified on both sides — Input and output rates are verified for both models.
- 5/5 core specs verified on both sides — All core specifications verified for both models.
- 1 shared named benchmark with differing scores — Measured on: Terminal-Bench 2.1.
- Verified within the last 90 days — Newest catalog check was 6 days ago.
- Both models carry source citations — Each side has at least two catalog sources on record.
Source receipts
Each catalog figure was checked against the provider or an independent second source on the date shown.
- Muse Spark 1.3: Meta — Introducing Muse Spark 1.3 (accessed 2026-09-03)
- Muse Spark 1.3: Meta developer — Muse Spark pricing ($1.25/$4.25) (accessed 2026-09-03)
- Muse Spark 1.3: VentureBeat — Muse Spark 1.3 best results need a config developers can't broadly use yet (accessed 2026-09-03)
- Muse Spark 1.3: HaiMaker — muse-spark-1.3 (131,072 max output tokens) (accessed 2026-09-05)
- Muse Spark 1.3: Promptfoo — Meta provider docs (muse-spark-1.3 max_tokens ceiling) (accessed 2026-09-05)
- Qwen 3.8 Max: Qwen — Qwen 3.8 Max (accessed 2026-08-29)
- Qwen 3.8 Max: Alibaba Cloud Model Studio (accessed 2026-08-29)
Related comparisons
- Claude Fable 5.1 vs Muse Spark 1.3
- Claude Fable 5.1 vs Qwen 3.8 Max
- Claude Fable 5 vs Muse Spark 1.3
- Claude Fable 5 vs Qwen 3.8 Max
- Claude Mythos 5.1 vs Muse Spark 1.3
- Claude Mythos 5.1 vs Qwen 3.8 Max
Diving deeper on one model? Muse Spark 1.3 · Qwen 3.8 Max
Common questions
Muse Spark 1.3 vs Qwen 3.8 Max
Answered from the verified figures on this page rather than general guidance.
Is Muse Spark 1.3 or Qwen 3.8 Max cheaper for input?
Muse Spark 1.3 is cheaper at $1.25 per million input tokens, against $2 for Qwen 3.8 Max — roughly 1.6× the price. Output tokens usually dominate a real bill, so weigh the output rate more heavily than the input rate. These are base rates; Muse Spark 1.3 has tiered pricing: $1.25/$4.25 per MTok on Meta's Model API (cached input $0.15); a $0.10/$0.20 'contributor' tier applies when your data is used for training. Third-party reports put list at $1.50/$7.50 from 2027-01-01 — confirm before long-term commits. Qwen 3.8 Max has tiered pricing: $2/$6 per MTok (Alibaba Cloud Model Studio). Weights announced open, verify they have dropped.
Is Muse Spark 1.3 or Qwen 3.8 Max cheaper for output?
Muse Spark 1.3 is cheaper at $4.25 per million output tokens, against $6 for Qwen 3.8 Max — roughly 1.4× the price. Output tokens usually dominate a real bill, so weigh the output rate more heavily than the input rate. These are base rates; Muse Spark 1.3 has tiered pricing: $1.25/$4.25 per MTok on Meta's Model API (cached input $0.15); a $0.10/$0.20 'contributor' tier applies when your data is used for training. Third-party reports put list at $1.50/$7.50 from 2027-01-01 — confirm before long-term commits. Qwen 3.8 Max has tiered pricing: $2/$6 per MTok (Alibaba Cloud Model Studio). Weights announced open, verify they have dropped.
Which has the larger context window, Muse Spark 1.3 or Qwen 3.8 Max?
Muse Spark 1.3 accepts 1M tokens against 991K for Qwen 3.8 Max. This only matters if you routinely send very long documents or large codebases.
Do Muse Spark 1.3 and Qwen 3.8 Max support the same reasoning levels?
Muse Spark 1.3 exposes low, medium, high, while Qwen 3.8 Max exposes low, high, max.
Should I use Muse Spark 1.3 or Qwen 3.8 Max?
Both sit in the frontier tier, so the choice usually comes down to price and context rather than capability. Muse Spark 1.3 suits long-horizon coding agents; Qwen 3.8 Max suits open-weight frontier work.
Can I self-host Muse Spark 1.3 or Qwen 3.8 Max?
Qwen 3.8 Max publishes open weights, but self-hosting and third-party provider access depend on the licence, hardware, serving support, and availability. Muse Spark 1.3 is a closed model whose supported access paths are controlled by its provider.
Next step
Choosing between them
The verified figures that separate this pair, computed from the catalog rather than restated boilerplate.
- Input price: Muse Spark 1.3 costs $1.25 per 1M tokens versus $2 for Qwen 3.8 Max — a 1.6x difference at the headline tier.
- Context: Muse Spark 1.3 takes 1M against 991K for Qwen 3.8 Max — only decisive if your prompts approach the smaller window.
- Deployment: Qwen 3.8 Max publishes weights you can self-host; the other is API-only.
Choosing a harness rather than a model — Cursor, Copilot, Claude Code, Windsurf, or Lovable? Compare agentic harnesses · Latest releases.