MCP Atlas
A benchmark for tool-calling over Model Context Protocol integrations and external tools.
Top models on MCP Atlas — October 27, 2026
As of October 27, 2026, Muse Spark Open leads the MCP Atlas leaderboard with 89.2% , followed by Muse Spark 1.1 (88.1%) and Claude Opus 5 (85.8%).
Muse Spark Open
Meta
Muse Spark 1.1
Meta
Claude Opus 5
Anthropic
52 modelsAgentic4% of Agentic reference weightCurrentUpdated October 27, 2026
Leaderboard (52 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Muse Spark OpenMeta | 89.2% | Not reported | Open |
| 2 | Muse Spark 1.1Meta | 88.1% | Not reported | Closed |
| 3 | Claude Opus 5Anthropic | 85.8% | Not reported | Closed |
| 4 | Step 5 PreviewStepFun | 85.6% | Not reported | Open |
| 5 | Kimi K3Moonshot AI | 84.2% | Not reported | Pending |
| 6 | Hy4 previewTencent | 83.7% | Not reported | Open |
| 7 | Gemini 3.5 FlashGoogle | 83.6% | Not reported | Closed |
| 8 | Ling 3.1 FlashInclusionAI | 83% | Not reported | Open |
| 9 | Claude Opus 4.8Anthropic | 82.2% | Not reported | Closed |
| 10 | Beam 1.0 501B-A23B FP8Reflection AI | 81.5% | Not reported | Open |
| 11 | Beam 1.0 501B-A23BReflection AI | 80.8% | Not reported | Open |
| 12 | Beam 1.0 501B-A23B NVFP4Reflection AI | 80.5% | Not reported | Open |
| 13 | Ornith-1.5-397BOrnith AI | 80% | Not reported | Open |
| 14 | Inkling-SmallThinking Machines Lab | 79.6% | Not reported | Open |
| 15 | Claude Opus 4.7 (Adaptive)Anthropic | 77.3% | Not reported | Closed |
| 16 | GLM-5.2Z.AI | 76.8% | Not reported | Open |
| 17 | Qwen3.7 MaxAlibaba | 76.4% | Not reported | Closed |
| 18 | Kimi K2.7 CodeMoonshot AI | 76% | Not reported | Open |
| 19 | Ling 3.0 Flash SanteInclusionAI | 75.6% | Not reported | Open |
| 20 | Muse Glimmer 30BMeta | 75.5% | Not reported | Open |
| 21 | GPT-5.5OpenAI | 75.3% | Not reported | Closed |
| 22 | MiniMax M3MiniMax | 74.2% | Not reported | Open |
| 23 | DeepSeek V4 Pro (High)DeepSeek | 74.2% | Not reported | Open |
| 24 | InklingThinking Machines Lab | 74.1% | Not reported | Open |
| 25 | DeepSeek V4 Pro 0813DeepSeek | 73.6% | Not reported | Open |
| 26 | Qwen3.7 PlusAlibaba | 73.2% | Not reported | Closed |
| 27 | GLM-5.1Z.AI | 71.8% | Not reported | Open |
| 28 | GPT-5.4OpenAI | 70.6% | Not reported | Closed |
| 29 | Ornith-1.5-35B-A3BOrnith AI | 70.2% | Not reported | Open |
| 30 | DeepSeek V4 ProDeepSeek | 69.4% | Not reported | Open |
| 31 | DeepSeek V4 Flash 0731DeepSeek | 69% | Not reported | Open |
| 32 | DeepSeek V4 Flash (High)DeepSeek | 67.4% | Not reported | Open |
| 33 | Ling 3.0 FlashInclusionAI | 65.5% | Not reported | Open |
| 34 | DeepSeek V4 FlashDeepSeek | 64% | Not reported | Open |
| 35 | Qwen3.6-35B-A3BAlibaba | 62.8% | Not reported | Open |
| 36 | Spark-X2.6-4BSparkLLM | 62.4% | Not reported | Open |
| 37 | Solar Pro 4Upstage | 61.4% | Not reported | Closed |
| 38 | Solar Open 2Upstage | 58.2% | Not reported | Open |
| 39 | GPT-5.4 miniOpenAI | 57.7% | Not reported | Closed |
| 40 | GPT-5.4 nanoOpenAI | 56.1% | Not reported | Closed |
| 41 | Kimi K2.6Moonshot AI | 55.9% | Not reported | Open |
| 42 | Nemotron 3.5 Bolt 74B-A9B BF16NVIDIA | 55.6% | Not reported | Open |
| 43 | Ornith-1.5-9BOrnith AI | 54.2% | Not reported | Open |
| 44 | Nemotron 3.5 Bolt 74B-A9BNVIDIA | 52.2% | Not reported | Open |
| 45 | Qwen3.6 PlusAlibaba | 48.2% | Not reported | Closed |
| 46 | LLaDA2.2-flashInclusionAI | 46.2% | Not reported | Open |
| 47 | Qwen3.5 397BAlibaba | 46.1% | Not reported | Open |
| 48 | LongCat-Flash-Lite-SparseMeituan | 45.6% | Not reported | Open |
| 49 | Claude Opus 4.5Anthropic | 42.3% | Not reported | Closed |
| 50 | LFM3-1.7BLiquidAI | 32.6% | Not reported | Open |
| 51 | GLM-5Z.AI | 31.1% | Not reported | Open |
| 52 | Kimi K2.5Moonshot AI | 29.5% | Not reported | Open |
According to BenchLM.ai, Muse Spark Open leads the MCP Atlas benchmark with a score of 89.2%, followed by Muse Spark 1.1 (88.1%) and Claude Opus 5 (85.8%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.
52 models have been evaluated on MCP Atlas. The benchmark falls in the Agentic category. BenchAlign v5.8 gives MCP Atlas 4% of the Agentic reference weight, so it moves the Agentic leaderboard and the overall ranking. Reference weights are relative weights in the calibrated model, not fixed shares of a score.
About MCP Atlas
Year
2026
Tasks
Tool-integrated agent tasks
Format
Interactive tool-calling evaluation
Difficulty
Advanced tool use
OpenAI reports MCP Atlas as a tool-use benchmark that measures how well models work with MCP-backed systems and external tools.
Freshness and provenance
Version
MCP Atlas 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does MCP Atlas measure?
A benchmark for tool-calling over Model Context Protocol integrations and external tools.
Which model scores highest on MCP Atlas?
Muse Spark Open by Meta currently leads with a score of 89.2% on MCP Atlas.
How many models are evaluated on MCP Atlas?
52 AI models have published results on MCP Atlas in the BenchLM catalog.
Compare top models on MCP Atlas
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.