whatmodel.sh / GPT-3 davinci (API base model)
GPT-3 davinci (API base model)
↗ share
+ contribute
GPT-3 davinci (API base model) is a closed-weights language model from OpenAI, released June 2020. Licence: proprietary.
Catalogue identity notes
The original OpenAI API base model 'davinci' (2020-2023), retired by OpenAI in January 2024. Not the 2023 replacement babbage-002/davinci-002 and not the InstructGPT 'text-' models. OpenAI never officially published its parameter count, so params_b is omitted.
$ how it compares
The LLMs ranked either side of GPT-3 davinci (API base model), out of 828 with a score, then more from OpenAI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 712 |  StarCoder2 3B | 33±1 | local | ||
| 713 |  Granite 3B Code Base (2K) | 33±1 | local | ||
| 714 | GPT-3 davinci (API base model) | 33±1 | comparison only | ||
| 715 | StarCoderBase 15.5B | 33±1 | local | ||
| 716 |  Vicuna 7B v1.5 | 33±1 | local | ||
| more from OpenAI | |||||
| 1 | GPT-6 Astra | 87±1 | api $20.00/M | ||
| 3 | GPT-6.1 Sol | 85±1 | api $4.00/M | ||
| 7 | GPT-6 Sol | 83±1 | api $4.00/M | ||
$ what it's good at
other tests9 not in a task · 99% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| WinoGrande +1 | 72.0% ✎ | ▄·▄·▄▄▄··█·▄ beats 7/7 | 22% |
| ARC Challenge | 43.5% ✎ | ▄····▄···█▄▄ beats 5/5 | 21% |
| HellaSwag +1 | 78.0% ✎ | ▄·█·█··▄▄··▄ beats 7/7 | 17% |
| TriviaQA | 40.9% ✎ | █···█·███··█ beats 5/5 | 11% |
| OpenBookQA | 33.6% ✎ | ▅·······█··▃ beats 5/5 | 10% |
| LAMBADA (OpenAI version), last-word accuracy +1 | 75.0% ✎ | █·▄··█··█··▄ beats 7/7 | 6.7% |
| PIQA +1 | 80.0% ✎ | █··▃··▃▃·▃·▃ beats 7/7 | 5.5% |
| ARC Easy | 76.2% ✎ | ▄···▄··█·▄·▄ beats 5/5 | 4.6% |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · GPT-NeoX-20B: An Open-Source… 75%, EleutherAI GPT-J-6B model car… 25%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 GPT-NeoX-20B: An Open-Source Autoregressive Language Model — authors' own runs | 9 | 75% |
 EleutherAI GPT-J-6B model card — EleutherAI lm-evaluation-harness runs | 4 | 25% |
| prior · pull toward the middle | 1% | |
| weighted average | = 33.1 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json