Excel Modeling Benchmark v1.1 (EMB v1.1)
We show this table for reference; we do not rank on it.
Excel financial modeling from a template or from scratch, with updated grading in v1.1.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Category-mean accuracy on EMB v1.1 — October 27, 2026
We mirror the published category-mean accuracy view for EMB v1.1. Claude Fable 5.1 leads the public snapshot at 76.73%, followed by Gemini 4 Carbon (76.47%) and Claude Sonnet 5.5 (75.90%). We do not use these results to rank models overall.
Claude Fable 5.1
Anthropic
Gemini 4 Carbon
high reasoning
Claude Sonnet 5.5
Anthropic
35 modelsAgenticCurrentDisplay onlyUpdated October 27, 2026
Category-mean accuracy table (35 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.1Anthropic | 76.73% | Not reported | Closed |
| 2 | Gemini 4 CarbonGooglehigh reasoning | 76.47% | Not reported | Closed |
| 3 | Claude Sonnet 5.5Anthropic | 75.90% | Not reported | Closed |
| 4 | Claude Opus 5.5Anthropic | 75.84% | Not reported | Closed |
| 5 | Muse Wonder FullMetamax reasoning | 75.00% | Not reported | Closed |
| 6 | Gemini 4 ArgonGooglehigh reasoning | 74.83% | Not reported | Closed |
| 7 | GPT-6.1 TerraOpenAImax reasoning | 73.33% | Not reported | Closed |
| 8 | Gemini 3.8 FlashGooglehigh reasoning | 72.87% | Not reported | Closed |
| 9 | GPT-6 AstraOpenAImax reasoning | 72.68% | Not reported | Closed |
| 10 | Muse WonderMetamax reasoning | 72.50% | Not reported | Closed |
| 11 | Qwen3.8 Max InternalAlibaba | 71.98% | Not reported | Unknown |
| 12 | Grok 4.7xAIxhigh reasoning | 71.50% | Not reported | Closed |
| 13 | Kimi K3Moonshot AImax reasoning | 71.26% | Not reported | Open |
| 14 | GPT-6.1 SolOpenAImax reasoning | 71.09% | Not reported | Closed |
| 15 | Kimi K3.1 ProMoonshot AImax reasoning | 70.00% | Not reported | Pending |
| 16 | GPT-6 LunaOpenAImax reasoning | 69.95% | Not reported | Closed |
| 17 | Step 5 PreviewStepFun | 69.55% | Not reported | Open |
| 18 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 69.17% | Not reported | Open |
| 19 | GLM-5.3Z.AImax reasoning | 69.11% | Not reported | Open |
| 20 | Mistral Large 4Mistralhigh reasoning | 68.97% | Not reported | Open |
| 21 | Hy4 previewTencent | 68.95% | Not reported | Open |
| 22 | MiMo-V2.6-FlashXiaomi | 68.88% | Not reported | Open |
| 23 | MiniMax M3.1 FlashMiniMax | 67.84% | Not reported | Closed |
| 24 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 67.76% | Not reported | Open |
| 25 | MiMo-V2.6-ProXiaomi | 67.35% | Not reported | Open |
| 26 | GLM-5.3-FlashZ.AImax reasoning | 66.25% | Not reported | Open |
| 27 | Qwen3.8 MaxAlibaba | 65.99% | Not reported | Open |
| 28 | Claude Haiku 5.5Anthropic | 65.48% | Not reported | Closed |
| 29 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 65.13% | Not reported | Open |
| 30 | Beam 1.0Reflection | 64.73% | Not reported | Unknown |
| 31 | Muse Spark OpenMetaxhigh reasoning | 61.98% | Not reported | Open |
| 32 | MiniMax M3MiniMax | 59.75% | Not reported | Open |
| 33 | Mercury 3Inceptionhigh reasoning | 57.88% | Not reported | Closed |
| 34 | InklingThinking Machines Lab0.99 reasoning | 40.97% | Not reported | Open |
| 35 | Inkling-SmallThinking Machines Lab0.99 reasoning | 40.82% | Not reported | Open |
How EMB v1.1 is shown here
BenchLM mirrors the public Vals AI EMB v1.1 leaderboard captured from https://www.vals.ai/benchmarks/emb-v1-1 and updated by Vals on October 27, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.
EMB v1.1 is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.
Snapshot
The published EMB v1.1 snapshot places Claude Fable 5.1 first at 76.73%. The third row is 0.83 points behind. The broader top-10 range is 4.22 points, so many of the published results sit in a relatively narrow band.
35 models have been evaluated on EMB v1.1. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. EMB v1.1 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About EMB v1.1
Year
2026
Tasks
Excel-based financial modeling tasks
Format
Equally weighted mean of seven category scores
Difficulty
Professional finance modeling
Version 1.1 switches the Scratch judge to GPT-6 Luna and fixes Template sheet-name matching. Accuracy averages tasks within seven categories, then weights the categories equally. Scores are not directly comparable with v1.0. Partial credit can include incomplete work; the table stays display only.
Freshness and provenance
Version
EMB v1.1 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 EMB v1.1 measure?
Excel financial modeling from a template or from scratch, with updated grading in v1.1.
Which model leads the published EMB v1.1 snapshot?
Claude Fable 5.1 currently leads the published EMB v1.1 snapshot with 76.73% category-mean accuracy. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on EMB v1.1?
The October 27, 2026 snapshot contains 35 AI models.
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.