Leaderboard
Ratings from blind comparisons in Slovenian. A longer line means a less certain rating.
| Ranked models (5)Models with at least 30 counted votes. | |||||
|---|---|---|---|---|---|
| GaMS3-12B-InstructCJVT | 1,072 | ±61 | 371 | 62.6% | |
| gpt-5OpenAI | 1,066 | ±77 | 96 | 55.8% | |
| GaMS-9B-Instruct-NemotronCJVT | 1,063 | ±65 | 218 | 58.3% | |
| gpt-5-miniOpenAI | 1,032 | ±77 | 92 | 49.3% | |
| gpt-oss-20bOpenAI | 868 | ±74 | 130 | 20.8% | |
| New models (12)Still collecting votes (fewer than 30); the rating may still change a lot. | |||||
| deepseek-v4.1-flashDeepSeek | 1,248 | ±235 | 1 | 100.0% | |
| Qwen3.8-Flash-NextQwen | 1,208 | ±193 | 1 | 100.0% | |
| GaMS2-27B-InstructCJVT | 1,151 | ±141 | 23 | 76.2% | |
| gpt-5.6-lunaOpenAI | 1,147 | ±192 | 13 | 75.0% | |
| glm-5.3Z.ai | 1,061 | ±87 | 1 | 100.0% | |
| gemini-3.1-pro-previewGoogle | 1,059 | ±166 | 13 | 50.0% | |
| gemini-3.8-flashGoogle | 1,057 | ±130 | 25 | 54.5% | |
| gemma-4-31b-itGoogle | 1,043 | ±165 | 12 | 55.6% | |
| gemma-4-26b-a4b-itGoogle | 952 | ±187 | 11 | 28.6% | |
| gemini-3.5-flashGoogle | 895 | ±169 | 3 | 50.0% | |
| nemotron-3-ultra-550b-a55bNVIDIA | 868 | ±135 | 18 | 15.4% | |
| Qwen3.6-27BQwen | 843 | ±158 | 1 | 0.0% | |
| Inactive models (29)No longer in comparisons. | |||||
| gemini-3-pro-previewGoogle | 1,232 | ±154 | 27 | 79.2% | |
| gemini-2.5-proGoogle | 1,170 | ±68 | 255 | 78.0% | |
| gemini-2.5-flashGoogle | 1,100 | ±89 | 74 | 63.5% | |
| GaMS-27B-Instruct-NemotronCJVT | 1,076 | ±69 | 256 | 65.0% | |
| gemini-2.0-flashGoogle | 1,065 | ±74 | 151 | 62.1% | |
| Mistral-Large-3-675B-Instruct-2512Mistral AI | 1,061 | ±99 | 59 | 52.2% | |
| gemma-3-27b-itGoogle | 1,061 | ±69 | 184 | 64.3% | |
| kimi-k2-instructMoonshot AI | 1,060 | ±90 | 70 | 56.4% | |
| Qwen3-Next-80B-A3B-InstructQwen | 1,038 | ±89 | 51 | 54.1% | |
| gpt-4oOpenAI | 1,029 | ±64 | 256 | 54.6% | |
| gemini-1.5-proGoogle | 1,024 | ±76 | 135 | 59.0% | |
| gpt-oss-120bOpenAI | 1,008 | ±78 | 87 | 44.1% | |
| deepseek-r1DeepSeek | 1,005 | ±77 | 157 | 52.1% | |
| Ministral-3-14B-Instruct-2512Mistral AI | 997 | ±95 | 55 | 41.0% | |
| deepseek-r1-0528DeepSeek | 994 | ±88 | 63 | 50.0% | |
| llama-4-maverick-17b-128e-instructMeta | 986 | ±69 | 213 | 44.1% | |
| Zlatorog-12B-Instruct-BetazID4si | 968 | ±69 | 191 | 42.7% | |
| gemma-3-12b-itGoogle | 962 | ±70 | 180 | 41.4% | |
| Qwen3-235B-A22BQwen | 959 | ±190 | 7 | 25.0% | |
| llama-4-scout-17b-16e-instructMeta | 959 | ±72 | 175 | 42.7% | |
| EuroLLM-9B-InstructUTTER | 941 | ±78 | 136 | 37.6% | |
| llama-3.3-70b-instructMeta | 926 | ±69 | 205 | 33.5% | |
| GaMS-9B-InstructCJVT | 906 | ±73 | 157 | 34.5% | |
| deepseek-v3.2DeepSeek | 895 | ±203 | 9 | 14.3% | |
| GaMS-27B-InstructCJVT | 891 | ±80 | 114 | 30.0% | |
| Mistral-Small-24B-Instruct-2501Mistral AI | 836 | ±117 | 37 | 20.0% | |
| bielik-11b-v2.3-instructSpeakLeash | 815 | ±72 | 181 | 15.5% | |
| llama3-70b-instructMeta | 745 | ±182 | 20 | 15.8% | |
| nemotron-3.5-lightning-30b-a3bNVIDIA | 655 | ±125 | 10 | 0.0% | |
46 models
Ranked models (5)Models with at least 30 counted votes.
- GaMS3-12B-InstructCJVT1,072±61371 votes · 62.6% wins
- gpt-5OpenAI1,066±7796 votes · 55.8% wins
- GaMS-9B-Instruct-NemotronCJVT1,063±65218 votes · 58.3% wins
- gpt-5-miniOpenAI1,032±7792 votes · 49.3% wins
- gpt-oss-20bOpenAI868±74130 votes · 20.8% wins
New models (12)Still collecting votes (fewer than 30); the rating may still change a lot.
- deepseek-v4.1-flashDeepSeek · new1,248±2351 vote · 100.0% wins
- Qwen3.8-Flash-NextQwen · new1,208±1931 vote · 100.0% wins
- GaMS2-27B-InstructCJVT · new1,151±14123 votes · 76.2% wins
- gpt-5.6-lunaOpenAI · new1,147±19213 votes · 75.0% wins
- glm-5.3Z.ai · new1,061±871 vote · 100.0% wins
- gemini-3.1-pro-previewGoogle · new1,059±16613 votes · 50.0% wins
- gemini-3.8-flashGoogle · new1,057±13025 votes · 54.5% wins
- gemma-4-31b-itGoogle · new1,043±16512 votes · 55.6% wins
- gemma-4-26b-a4b-itGoogle · new952±18711 votes · 28.6% wins
- gemini-3.5-flashGoogle · new895±1693 votes · 50.0% wins
- nemotron-3-ultra-550b-a55bNVIDIA · new868±13518 votes · 15.4% wins
- Qwen3.6-27BQwen · new843±1581 vote · 0.0% wins
Inactive models (29)No longer in comparisons.
- gemini-3-pro-previewinactiveGoogle1,232±15427 votes · 79.2% wins
- gemini-2.5-proinactiveGoogle1,170±68255 votes · 78.0% wins
- gemini-2.5-flashinactiveGoogle1,100±8974 votes · 63.5% wins
- GaMS-27B-Instruct-NemotroninactiveCJVT1,076±69256 votes · 65.0% wins
- gemini-2.0-flashinactiveGoogle1,065±74151 votes · 62.1% wins
- Mistral-Large-3-675B-Instruct-2512inactiveMistral AI1,061±9959 votes · 52.2% wins
- gemma-3-27b-itinactiveGoogle1,061±69184 votes · 64.3% wins
- kimi-k2-instructinactiveMoonshot AI1,060±9070 votes · 56.4% wins
- Qwen3-Next-80B-A3B-InstructinactiveQwen1,038±8951 votes · 54.1% wins
- gpt-4oinactiveOpenAI1,029±64256 votes · 54.6% wins
- gemini-1.5-proinactiveGoogle1,024±76135 votes · 59.0% wins
- gpt-oss-120binactiveOpenAI1,008±7887 votes · 44.1% wins
- deepseek-r1inactiveDeepSeek1,005±77157 votes · 52.1% wins
- Ministral-3-14B-Instruct-2512inactiveMistral AI997±9555 votes · 41.0% wins
- deepseek-r1-0528inactiveDeepSeek994±8863 votes · 50.0% wins
- llama-4-maverick-17b-128e-instructinactiveMeta986±69213 votes · 44.1% wins
- Zlatorog-12B-Instruct-BetainactivezID4si968±69191 votes · 42.7% wins
- gemma-3-12b-itinactiveGoogle962±70180 votes · 41.4% wins
- Qwen3-235B-A22BinactiveQwen959±1907 votes · 25.0% wins
- llama-4-scout-17b-16e-instructinactiveMeta959±72175 votes · 42.7% wins
- EuroLLM-9B-InstructinactiveUTTER941±78136 votes · 37.6% wins
- llama-3.3-70b-instructinactiveMeta926±69205 votes · 33.5% wins
- GaMS-9B-InstructinactiveCJVT906±73157 votes · 34.5% wins
- deepseek-v3.2inactiveDeepSeek895±2039 votes · 14.3% wins
- GaMS-27B-InstructinactiveCJVT891±80114 votes · 30.0% wins
- Mistral-Small-24B-Instruct-2501inactiveMistral AI836±11737 votes · 20.0% wins
- bielik-11b-v2.3-instructinactiveSpeakLeash815±72181 votes · 15.5% wins
- llama3-70b-instructinactiveMeta745±18220 votes · 15.8% wins
- nemotron-3.5-lightning-30b-a3binactiveNVIDIA655±12510 votes · 0.0% wins
How to read ±95%. The vertical mark is the model's rating; the line is its 95% confidence interval, the range of ratings still compatible with the votes so far. More votes make the line shorter. When two lines overlap, the votes do not yet tell the two models apart, which does not mean they are equally good.
A new model does not yet have enough votes for a reliable rank (at least 30), so it is shown with its rating but without a rank, below the ranked models. An inactive model no longer takes part in new conversations and gets no new votes, so it has no rank and is listed at the bottom of the leaderboard, below the active models. Untick “Also show inactive models” to hide it. Its rating can still shift a little as the ratings of the models it was compared with change. Win rate counts only votes with a winner: “It’s a tie” and “Both are bad” are left out of it.
Missing a model? Suggest it.