text model comparison
Inkling 2 Small vs Muse Spark Open
Inkling 2 Small and Muse Spark Open are both current text models, from Thinking Machines Lab and Meta. In Thinking Machines Lab's Inkling 2 Small model card, Inkling 2 Small scores 72.1% on DeepSWE v1.1 and Muse Spark Open 60.0%. Muse Spark Open has the larger context window, 1,000,000 vs 262,144 tokens.
- Longer context window
- Muse Spark Open1M tokens against 262K tokens
- Newer release
- Inkling 2 SmallNov 2026 against Oct 2026
Data verified November 6, 2026 · Dataset 2026-11-06
Side by side
Inkling 2 Small and Muse Spark Open, line by line
Prices in effect today, compared at the same tier wherever both list one. The better value in each row is highlighted.
| Fact | Inkling 2 Small | Muse Spark Open |
|---|---|---|
| Price | ||
| Price per 1M input tokens | $0.30 | Not listed |
| Price per 1M output tokens | $1 | Not listed |
| Independent benchmarks | ||
| Arena WebDev | Not measured | 1526 Elo (#36 of 46) |
| GPQA Diamond | Not measured | 88.6% (#34 of 41) |
| FrontierMath Tiers 1-3 (v2) | Not measured | 69.8% (#24 of 41) |
| SimpleQA Verified | Not measured | 56.6% (#16 of 38) |
| Specs | ||
| Provider | Thinking Machines Lab | Meta |
| Release date | November 5, 2026 | October 15, 2026 |
| Access | API only | open weights |
| Context window | 262,144 tokens | 1,000,000 tokens |
| Max output | Not reported | Not reported |
| Input | text, image | text, image, video, audio |
| Output | text | text |
No independent benchmark has scored both models yet. Each rank counts the current models scored on that benchmark.
Provider results
What the labs report
Results tables a lab published with one of its models, where it lists both Inkling 2 Small and Muse Spark Open. These are the lab's own figures, not independent measurements, and each table is read on its own: labs run benchmarks with different harnesses and budgets.
Thinking Machines Lab, Inkling 2 Small model card, November 5, 2026Thinking Machines Lab · 2026-11-06
| Benchmark | Inkling 2 Small | Muse Spark Open |
|---|---|---|
| DeepSWE v1.1 | 72.1% (better) | 60.0% |
| Vibe Code Bench v1.1 | 81.6% (better) | 76.2% |
| SWE-bench Multilingual | 81.2% (better) | 77.5% |
| SWE-bench Pro v2hard set | 86.3% (better) | 82.4% |
| SWE-bench Pro v2full set | 97.7% (better) | 95.0% |
| SWE-bench Verified | 82.2% (better) | 80.0% |
| MCP Atlasfull set | 85.4% | 89.2% (better) |
| MCP Atlaspublic set | 85.0% | 87.5% (better) |
| Finance Agent v2.1 | 67.5% (better) | 60.7% |
| Excel Modeling Benchmark v1.1 | 69.5% (better) | 62.0% |
| Tax Agent Bench | 65.8% (better) | 52.1% |
| APEX-Accounting v1.1pass@1 | 14.5% (better) | 13.8% |
| APEX-Accounting v1.1mean score | 62.0% (better) | 58.0% |
| Diligence Benchbest harness | 58.1% | 59.0% (better) |
| Harvey LAB v1.1 | 21.7% | 25.0% (better) |
| Legal Research Bench | 46.6% (better) | 38.5% |
| BabyVision | 68.6% (better) | 44.3% |
| BabyVisionwith Python | 93.7% (better) | 80.2% |
| CharXiv (RQ) | 85.6% (better) | 85.4% |
| CharXiv (RQ)with Python | 87.7% | 88.7% (better) |
| Chartography | 36.7% (better) | 35.7% |
| Chartographywith Python | 80.4% (better) | 63.2% |
| MathVision | 92.8% (better) | 87.7% |
| SAGE v1.1 | 88.6% (better) | 85.8% |
| GPQA-Diamond | 93.0% (better) | 89.0% |
| Apex 2025 | 63.5% (better) | 32.3% |
| HMMT Feb 2026 | 95.8% (better) | 93.9% |
| LongBench v2 | 70.6% (better) | 68.7% |
| ComplexConstraints v1.0 | 44.5% (better) | 41.3% |
| PatchEval | 79.6% (better) | 77.8% |
Thinking Machines' table from the Inkling 2 Small model card, run at effort 0.99 (xhigh) with a 1M-token trajectory limit; where a benchmark's owner publishes scores (Datacurve, Scale AI, Surge, Mercor, Thoughtful, Vals), Thinking Machines uses them. DeepSeek-V4.1-Flash is the 1017 build. Left out here: the Artificial Analysis and Vals indexes, the rows Artificial Analysis ran (FrontierCode, Terminal-Bench, GDP.pdf, MMMU-Pro, HLE, CritPt, AA-LCR, AA-Omniscience, AutomationBench, EnterpriseOps-Gym, Tau 3 Banking), Arena and DesignArena ratings, PMPP-Hard (not a percentage), the safety rows, and JobBench, DAYJOB and Humanity's Sixth Sense, which we could not confirm are public.
Charts
Against every current text model
Where Inkling 2 Small and Muse Spark Open land among the current models of their task, on every independent benchmark, on price and on context.
Where each stands
Only Muse Spark Open has a current Arena WebDev score: #36 of 46. Each bar is the share of current models it beats.
Independent benchmarks
Arena WebDev
Inkling 2 Small: Not scored
Muse Spark Open:#36of 46
1,526 Elo · ahead of 22%
GPQA Diamond
Inkling 2 Small: Not scored
Muse Spark Open:#34of 41
88.6% · ahead of 18%
FrontierMath Tiers 1-3 (v2)
Inkling 2 Small: Not scored
Muse Spark Open:#24of 41
69.8% · ahead of 43%
SimpleQA Verified
Inkling 2 Small: Not scored
Muse Spark Open:#16of 38
56.6% · ahead of 59%
Price and context
Price per 1M input tokens
Inkling 2 Small:#11of 58
$0.30 · cheaper than 75%
Muse Spark Open: Not listed
Price per 1M output tokens
Inkling 2 Small:#10of 58
$1 · cheaper than 84%
Muse Spark Open: Not listed
Context window
Inkling 2 Small:#57of 67
262K tokens · longer than 12%
Muse Spark Open:#27of 67
1M tokens · longer than 24%
Cost
What each costs
Example workloads at each provider's listed rates, then every other rate either one lists. The cheaper figure in each row is highlighted.
| Workload or rate | Inkling 2 Small | Muse Spark Open |
|---|---|---|
| Other listed rates | ||
| 1M input tokens, cached | $0.03 | Not listed |
Your own bill will differ. These figures use list prices and example workloads. The AI Model Cost Audit reviews your real usage, shortlists models that score as well for each task for less, and estimates the savings.
Launch price $100, regular price $150
Verdict
When to choose each
Inkling 2 Small
Where it leads
- A cached-input rate, $0.03 per 1M tokens
Best for
- Agentic coding and office, finance and legal work at a low per-token price
- Teams that want to post-train their own variant on Tinker
Watch out for
- Closed weights for now: Thinking Machines says it will publish the full weights under Apache 2.0 the week after launch, once external safety testing ends; until then Tinker is the only way to use it
- Tinker serves it with a 64K or 256K context, although the model card lists up to 1M tokens
- In its lab's own table DeepSeek-V4.1-Flash scores higher on DeepSWE v1.1 (77.0% against 72.1%) and Claude Haiku 5.5 on Vibe Code Bench (90.4% against 81.6%); Thinking Machines also says it is less tuned for chat, and no independent benchmark score is recorded here yet
Muse Spark Open
Where it leads
- Larger context window 1,000,000 vs 262,144 tokens
- Accepts video and audio input
- Open weights you can run yourself
Best for
- Self-hosted coding and agent work on owned infrastructure
- Long-context multimodal analysis where data must stay in-house
Watch out for
- The Muse Model License limits commercial use; it is not an open-source licence
- Meta does not serve it on its own Model API, so hosted prices come from third parties
- Self-hosting needs at least 700GB of GPU memory in NVFP4 or 2TB in BF16
From benchmark to production
Building on Inkling 2 Small or Muse Spark Open?
Public scores rarely hold on your own data. We test both models on your real tasks, then build the agent or pipeline around the one that wins.
AI Model Cost AuditLaunch price $100, regular price $150What the audit includes
FAQ
Inkling 2 Small vs Muse Spark Open FAQ
Do Inkling 2 Small and Muse Spark Open discount cached input?
Inkling 2 Small lists a cached-input rate of $0.03 per 1M tokens, 90% below its standard input rate. No cached-input rate is listed for Muse Spark Open.
How do Inkling 2 Small and Muse Spark Open compare in Thinking Machines Lab's own results?
In the Inkling 2 Small model card (November 5, 2026), Inkling 2 Small leads on DeepSWE v1.1, 72.1% to 60.0%; Inkling 2 Small leads on Vibe Code Bench v1.1, 81.6% to 76.2%; Inkling 2 Small leads on SWE-bench Multilingual, 81.2% to 77.5%. These are Thinking Machines Lab's figures, not independent measurements.
Which has the larger context window, Inkling 2 Small or Muse Spark Open?
Muse Spark Open has the larger context window, 1,000,000 vs 262,144 tokens.
How do Inkling 2 Small and Muse Spark Open differ in features?
Inkling 2 Small takes text and image input; Muse Spark Open takes text, image, video and audio.
Can I run Inkling 2 Small or Muse Spark Open myself?
Muse Spark Open is released as open weights you can run yourself; Inkling 2 Small is available through its provider's API only.
Further reading
More on Inkling 2 Small and Muse Spark Open
Related comparisons
- Claude Haiku 5.5 vs Inkling 2 Small
- Muse Spark 1.3 vs Muse Spark Open
- DeepSeek-V4.1-Flash vs Inkling 2 Small
- Muse Spark Open vs Muse Wonder
- Gemini 3.5 Flash-Lite vs Inkling 2 Small
- MiniMax M3.1 Flash vs Muse Spark Open
- Gemini 3.8 Flash vs Inkling 2 Small
- Beam 1.0 vs Muse Spark Open
- GLM-5.3-Flash vs Inkling 2 Small
- Muse Spark Open vs Nemotron 3.5 Super VL
- GLM-5.3 vs Inkling 2 Small
- GPT-6 Luna vs Inkling 2 Small
Production evaluation
Choosing between Inkling 2 Small and Muse Spark Open?
Public benchmarks narrow the shortlist. We run both on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.
- A head-to-head evaluation on your real workload
- Routing between models by cost and difficulty