# ExtractBench (LlamaIndex) leaderboard

Multimodal & Grounded benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/multimodal-grounded-extract-bench-llama-index

ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard). Every model the catalog carries a published ExtractBench (LlamaIndex) value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Multimodal & Grounded |
| Measure | Order-insensitive value F1 overall and by document length; word and page grounding F1; cost and latency |
| Tasks | 370 documents, 4,869 pages, 8 business domains, 67 document types, each with a JSON Schema |
| Difficulty | Short documents are near solved; long record lists and grounding separate systems |
| Published by | [ExtractBench: A Benchmark for Schema-Guided Enterprise Document Extraction](https://arxiv.org/abs/2607.29677) |

Scores schema-guided extraction from 370 enterprise documents and 4,869 pages on value F1, record completeness, word and page grounding and cost, split by short, medium and long documents.

## ExtractBench (LlamaIndex) ranking

24 models with a published ExtractBench (LlamaIndex) value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Order-insensitive value F1 overall and by document length; word and page grounding F1; cost and latency | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [Claude Sonnet 5.5](https://modelscale.dev/models/claude-sonnet-5-5) | Anthropic | 94.68 | Observed 2026-10-23 · source benchlm:models |
| 2 | [GPT-6 Sol](https://modelscale.dev/models/gpt-6-sol) | OpenAI | 94.57 | Observed 2026-10-23 · source benchlm:models |
| 3 | [GPT-5.6 Sol](https://modelscale.dev/models/gpt-5-6-sol) | OpenAI | 93.77 | Observed 2026-10-23 · source benchlm:models |
| 4 | [GPT-5.5](https://modelscale.dev/models/gpt-5-5) | OpenAI | 93.57 | Observed 2026-10-23 · source benchlm:models |
| 5 | [GPT-5.6 Terra](https://modelscale.dev/models/gpt-5-6-terra) | OpenAI | 92.26 | Observed 2026-10-23 · source benchlm:models |
| 6 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 91.91 | Observed 2026-10-23 · source benchlm:models |
| 7 | [GPT-5.6 Luna](https://modelscale.dev/models/gpt-5-6-luna) | OpenAI | 91.16 | Observed 2026-10-23 · source benchlm:models |
| 8 | [GPT-6 Luna](https://modelscale.dev/models/gpt-6-luna) | OpenAI | 91.01 | Observed 2026-10-23 · source benchlm:models |
| 9 | [Qwen3.8-Flash-Next](https://modelscale.dev/models/qwen3-8-flash-next) | Alibaba | 89.88 | Observed 2026-10-23 · source benchlm:models |
| 10 | [Qwen3.8-27B](https://modelscale.dev/models/qwen3-8-27b) | Alibaba | 89.75 | Observed 2026-10-23 · source benchlm:models |
| 11 | [Qwen3.6-35B-A3B](https://modelscale.dev/models/qwen3-6-35b-a3b) | Alibaba | 88.11 | Observed 2026-10-23 · source benchlm:models |
| 12 | [Qwen3.5-35B-A3B](https://modelscale.dev/models/qwen3-5-35b-a3b) | Alibaba | 88.00 | Observed 2026-10-23 · source benchlm:models |
| 13 | [DeepSeek V4.1 Flash 1017](https://modelscale.dev/models/deepseek-v4-1-flash) | DeepSeek | 87.11 | Observed 2026-10-23 · source benchlm:models |
| 14 | [Claude Opus 4.8](https://modelscale.dev/models/claude-opus-4-8) | Anthropic | 87.09 | Observed 2026-10-23 · source benchlm:models |
| 15 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 86.06 | Observed 2026-10-23 · source benchlm:models |
| 16 | [Kimi K3](https://modelscale.dev/models/kimi-3) | Moonshot AI | 83.17 | Observed 2026-10-23 · source benchlm:models |
| 17 | [GLM-5.3-Flash](https://modelscale.dev/models/glm-5-3-flash) | Z.AI | 80.75 | Observed 2026-10-23 · source benchlm:models |
| 18 | [Gemini 3.8 Flash](https://modelscale.dev/models/gemini-3-8-flash) | Google | 80.71 | Observed 2026-10-23 · source benchlm:models |
| 19 | [Gemini 3.5 Flash](https://modelscale.dev/models/gemini-3-5-flash) | Google | 79.84 | Observed 2026-10-23 · source benchlm:models |
| 20 | [GPT-5.4 nano](https://modelscale.dev/models/gpt-5-4-nano) | OpenAI | 74.90 | Observed 2026-10-23 · source benchlm:models |
| 21 | [Gemma 4 26B A4B](https://modelscale.dev/models/gemma-4-26b-a4b) | Google | 70.27 | Observed 2026-10-23 · source benchlm:models |
| 22 | [Gemma 4 E4B](https://modelscale.dev/models/gemma-4-e4b) | Google | 69.64 | Observed 2026-10-23 · source benchlm:models |
| 23 | [Gemma 4 E2B](https://modelscale.dev/models/gemma-4-e2b) | Google | 51.85 | Observed 2026-10-23 · source benchlm:models |
| 24 | [Gemma 4 12B](https://modelscale.dev/models/gemma-4-12b) | Google | 45.39 | Observed 2026-10-23 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.