LAMBADA
Megatron-Turing NLG 530B 87.2% leads since Oct 2021 Dropped Models predict the last word of narrative passages where humans can guess it only from the whole passage. Scored by accuracy.
In the authors’ words
We introduce LAMBADA, a dataset to evaluate the capabilities of computational models for text understanding by means of a word prediction task. LAMBADA is a collection of narrative passages sharing the characteristic that human subjects are able to guess their last word if they are exposed to the whole passage, but not if they only see the last sentence preceding the target word. To succeed on LAMBADA, computational models cannot simply rely on local context, but must be able to keep track of information in the broader discourse. We show that LAMBADA exemplifies a wide range of linguistic phenomena, and that none of several state-of-the-art language models reaches accuracy above 1% on this novel benchmark. We thus propose LAMBADA as a challenging test set, meant to encourage the development of new models capable of genuine understanding of broad context in natural language text.
- Level
- Everyday Most adults, without training
- Scored
- Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
- Answers
- Short answer A number, word or short string matched to a key.
Megatron-Turing NLG 530B leads with 87.2% among models released by Nov 2023
Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in Nov 2023, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 87.2 Megatron-Turing NLG 530B
- 86.4 InstructGPT 175B
- 79.8
Falcon-180B - 78.9
Llama 2-70B - 78.5 Inflection-1
- 77.9
PaLM (540B) - 77.7 LLaMA-65B
- 77.4 Chinchilla
- 77.3 Falcon-40B
- 77.2 LLaMA-33B
- 76.5 Llama 2-13B
- 75.2 LLaMA-13B
- 74.9 Falcon-7B
- 74.5 Gopher (280B)
- 74.0
Baichuan2-13B - 73.3 LLaMA-7B
- 73.3 Llama 2-7B
- 73.3 Baichuan 2-7B
- 71.8 internlm-20b
- 71.3
Stable Beluga 2 - 71.1
Qwen-14B - 70.0
MPT-7B - 67.9 Qwen-7B
- 67.0 internlm-7b
- 58.4 Qwen-1_8B
- 54.3 chatglm2-6b
- 1 Megatron-Turing NLG 530B 87.2%
- 2 InstructGPT 175B 86.4%
- 3 Falcon-180B 79.8%
- 4 Llama 2-70B 78.9%
- 5 Inflection-1 78.5%
- 6 PaLM (540B) 77.9%
- 7 LLaMA-65B 77.7%
- 8 Chinchilla 77.4%
- 9 Falcon-40B 77.3%
- 10 LLaMA-33B 77.2%
- 11 Llama 2-13B 76.5%
- 12 LLaMA-13B 75.2%
- 13 Falcon-7B 74.9%
- 14 Gopher (280B) 74.5%
- 15 Baichuan2-13B 74%
- 16 LLaMA-7B 73.3%
- 17 Llama 2-7B 73.3%
- 18 Baichuan 2-7B 73.3%
- 19 internlm-20b 71.8%
- 20 Stable Beluga 2 71.3%
- 21 Qwen-14B 71.1%
- 22 MPT-7B 70%
- 23 Qwen-7B 67.9%
- 24 internlm-7b 67%
- 25 Qwen-1_8B 58.4%
- 26 chatglm2-6b 54.3%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Megatron-Turing NLG 530B | 87.2% | Oct 2021 | Epoch AI | |
| 2 | InstructGPT 175B | 86.4% | Jan 2022 | Epoch AI | |
| 3 | Falcon-180B | 79.8% | Sep 2023 | Epoch AI | |
| 4 | Llama 2-70B | 78.9% | Jul 2023 | Epoch AI | |
| 5 | Inflection-1 | 78.5% | Jun 2023 | Epoch AI | |
| 6 | PaLM (540B) | 77.9% | Apr 2022 | Epoch AI | |
| 7 | LLaMA-65B | 77.7% | Feb 2023 | Epoch AI | |
| 8 | Chinchilla | 77.4% | Mar 2022 | Epoch AI | |
| 9 | Falcon-40B | 77.3% | Mar 2023 | Epoch AI | |
| 10 | LLaMA-33B | 77.2% | Feb 2023 | Epoch AI | |
| 11 | Llama 2-13B | 76.5% | Jul 2023 | Epoch AI | |
| 12 | LLaMA-13B | 75.2% | Feb 2023 | Epoch AI | |
| 13 | Falcon-7B | 74.9% | Apr 2023 | Epoch AI | |
| 14 | Gopher (280B) | 74.5% | Dec 2021 | Epoch AI | |
| 15 | Baichuan2-13B | 74% | Sep 2023 | Epoch AI | |
| 16 | LLaMA-7B | 73.3% | Feb 2023 | Epoch AI | |
| 17 | Llama 2-7B | 73.3% | Jul 2023 | Epoch AI | |
| 18 | Baichuan 2-7B | 73.3% | Sep 2023 | Epoch AI | |
| 19 | internlm-20b | 71.8% | Sep 2023 | Epoch AI | |
| 20 | Stable Beluga 2 | 71.3% | Jul 2023 | Epoch AI | |
| 21 | Qwen-14B | 71.1% | Sep 2023 | Epoch AI | |
| 22 | MPT-7B | 70% | May 2023 | Epoch AI | |
| 23 | Qwen-7B | 67.9% | Sep 2023 | Epoch AI | |
| 24 | internlm-7b | 67% | Jul 2023 | Epoch AI | |
| 25 | Qwen-1_8B | 58.4% | Nov 2023 | Epoch AI | |
| 26 | chatglm2-6b | 54.3% | Jun 2023 | Epoch AI |
Labs report scores for 1 model in their cards. We plot every source in the charts below.
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
The best score passed 85% in Oct 2021
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs dropped it from launch posts once the best score passed 85%
Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.
| Date | Launch post | Score | As the post writes it |
|---|---|---|---|
| 10 May 2023 | PaLM 2-L | 86.9 | LAMBADA, 1-shot |
More in Knowledge
- MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects.
Intern-S2-397B 89.8% - SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
- WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1.
Tencent Hy4 preview 83.9% - KMMLU-Pro In use Models answer multiple-choice questions taken from Korean National Professional Licensure exams. It tests professional knowledge used in Korean industry.
Solar Pro 4 79.2% - BIG-Bench Extra Hard In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is a harmonic mean of task accuracies. Gemma 4 31B IT 74.4%
- HLE-Verified In use Verified and revised version of Humanity's Last Exam (1,811 items) with uncertain items removed.
Seed2.0 Pro 73.6%