whatmodel.sh / Doubao Seed 2.0 Lite (260428)
Doubao Seed 2.0 Lite (260428)
↗ share
+ contribute
Doubao Seed 2.0 Lite (260428) is a closed-weights language model from ByteDance Seed, released April 2026. Licence: proprietary.
Catalogue identity notes
Exact Volcengine Ark model id 'doubao-seed-2-0-lite-260428' (LTS). ByteDance Seed's Seed2.0 page: 'Seed2.0 Lite was upgraded at the end of April' and reports 'Seed2.0 Lite(0428)' and 'Seed2.0 Lite(0215)' as separate columns. Distinct from seed-2-0-lite, which is the 'Seed2.0 Lite' of the arXiv Seed2.0 model card (2607.00248v1); that card names no dated checkpoint, so it may be 0215 or 0428. Matches Epoch version 'doubao-seed-2.0-lite-260428' (Doubao Seed 2.0 Lite (Apr 2026)).
Model information sources
Score within the LMArena (text) comparison set only.
$ how it compares
The LLMs ranked either side of Doubao Seed 2.0 Lite (260428), out of 897 with a score, then more from ByteDance Seed.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 98 | 69±2 | comparison only | |||
| 99 | 69±1 | api price unknown | |||
| 100 | 69±1 | api $0.20/M | |||
| 101 | 69±0 | api $1.09/M +local | |||
| 102 | 69±0 | api $2.15/M +local | |||
| more from ByteDance Seed | |||||
| 85 | 70±2 | api $1.78/M | |||
| 86 | 70±2 | api $0.89/M | |||
| 90 | 70±1 | api $0.95/M | |||
$ what it's good at
agent tasks█··········█ #2 of 2 · 23% of its score · 4 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Pro (split unspecified) lab | 46.6% ✎ | ▁·▁▁▁▁▁▄█▁▁▁ beats 27/64 | 8.9% |
| Terminal-Bench 2.0 lab | 43.3% ✎ | ▁▁▄▄▇█▇█▅▁▃▃ beats 16/38 | 5.8% |
| SWE-bench Multilingual lab | 66.6% ✎ | ▁··▂▂▃▆█▃▂▁▁ beats 29/69 | 5.3% |
| BrowseComp lab | 64.0% ✎ | ▁·▃▁▁▂▆▄█▃▂▄ beats 32/48 | 3.5% |
coding█··········█ #2 of 2 * · 14% of its score · 2 tests
| test | result | vs the others in its run | counts |
|---|---|---|---|
| SWE-bench Pro (split unspecified) lab | 46.6% ✎ | ▁·▁▁▁▁▁▄█▁▁▁ beats 27/64 | 8.9% |
| SWE-bench Multilingual lab | 66.6% ✎ | ▁··▂▂▃▆█▃▂▁▁ beats 29/69 | 5.3% |
math▁▁▃▅█▆▅█▄▁▁▁ #77 of 440 * · 4.8% of its score · 1 test
Beats Claude Sonnet 4.6, Qwen3.7 Plus, Claude Opus 4.5 here, though they rank higher overall. Trails INTELLECT-3, GPT-5 mini, Qwen3-Coder-Next (80B-A3B) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| OTIS mock AIME 2024–2025 | 91.1% ✎ | ▃▃▄▃▃▃▇▆█▄▃▆ beats 145/207 | 4.8% |
vision & documents█··········█ #1 of 2 · 13% of its score · 3 tests
Beats Doubao Seed 2.0 Pro (260215) here, though they rank higher overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv (reasoning questions) lab | 82.4% ✎ | ▂▄▃▄▅▅▅▅▅█▆▂ beats 89/105 | 4.6% |
| CharXiv descriptive questions lab | 94.5% ✎ | ▂▄▄▅▇▄██▇▃▃▂ beats 59/61 | 4.6% |
| MMMU Pro lab | 78.4% ✎ | ▁▂▁▂▅▃▃▆▂█▄▃ beats 48/55 | 3.8% |
web search█··········█ #2 of 2 * · 3.5% of its score · 1 test
| test | result | vs the others in its run | counts |
|---|---|---|---|
| BrowseComp lab | 64.0% ✎ | ▁·▃▁▁▂▆▄█▃▂▄ beats 32/48 | 3.5% |
other tests6 not in a task · 58% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GPQA Diamond +1 | 88.1% ✎ | ▁▁▁▁▂▃▅▄▄▆█▃ beats 163/221 | 26% |
| Chess Puzzles (Epoch) | 20.0% ✎ | █▁▁▁▁▂▄▂▃▃▂▁ beats 79/157 | 15% |
| SimpleQA Verified | 31.5% ✎ | ▁▂▂▂▅█▄▃▂··▁ beats 24/91 | 5.7% |
| Humanity's Last Exam lab | 25.7% ✎ | ▁▁▁▄▆▇▅█▆▁▂▁ beats 66/115 | 5.4% |
| MultiChallenge lab | 69.9% ✎ | ▁·▁·▁▅█▃▅▃█▃ beats 33/33 | 3.1% |
| SuperGPQA lab | 69.6% ✎ | ▂▁▂▂▅▆▃▇▇▇█▆ beats 47/48 | 3.1% |
$ where to run it
Through an API
api · Volcengine Ark · $0.20/M · $0.08906 in / $0.53436 out per 1M tokens · 256k context · docs.volcengine.com ↗ · price sourceUSD per 1M tokens, base tier (input up to 32K) from models.dev (CNY converted at 6.737012, 2026-08-26); higher tiers, audio input and cache excluded. Volcengine pricing page (observed 2026-11-12): doubao-seed-2.0-lite '[0, 32]K: 0.6 / … / 3.6' CNY, consistent. LTS model on Volcengine.
$ how the score was worked out · ByteDance Seed model page (la… 59%, Epoch AI — internal evaluatio… 40%
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 |
|---|---|---|
| 11 | 59% | |
| 4 | 40% | |
| prior · pull toward the middle | 1% | |
| weighted average | = 69.1 |
$ 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