Best AI models for web search for a 16GB graphics card
A leaderboard of the best AI models for web search for a 16GB graphics card (with 32GB of system RAM). The best one that runs on it right now is Mistral Small 3.1 24B (2503). Models are ranked on web research tests like BrowseComp, with search turned on. Models that run on it show the download that fits and the command to run it.
111 models · scores combine 12 benchmarks · updated how it works · find a leaderboard · benchmark your machine
Scores use reviewed web search evaluations only. Compare scores within this task. Models without task measurements are excluded.
Models for 32GB RAM · 16GB graphics card → ~16GB fast · reset
+ more
$ terminal
↗ share
detecting your OS…
Score vs API price
x
show
● open weights ◆ API only ─┘ best score for the price · x: USD per 1M tokens, 3 input : 1 output (log)
| # | model | task score↓ | where↓ | speed↓ | sources |
|---|
| # | model | task score↓ | where↓ | speed↓ | sources |
|---|---|---|---|---|---|
| 2 |  o3 | 92±2* | api $3.50/M |   | |
| 3 | o4-mini | 87±2* | api $1.93/M | | |
| 5 | GPT-4.1 | 79±3* | api $3.50/M | | |
| 7 | o1 (2024-12-17) | 75±3* | api $26.25/M | | |
| 8 | GPT-4.1 mini | 74±3* | api $0.70/M | | |
| 9 | GPT-4o (2024-08-06) | 74±3* | api $4.38/M | | |
| 11 | GPT-5.6 Sol | 67±5* | api $8.00/M |  | |
| 13 | o3-mini | 65±3* | api $1.93/M | | |
| 14 | GPT-6 Astra | 63±4*† | api $20.00/M | | |
| 15 | GPT-4o mini (2024-07-18) | 62±3* | api $0.26/M | | |
| 16 |  Mistral Small 3.1 24B (2503) · 4-bit | 62±3* | local (slow) | | |
| 17 |  Claude Opus 5.5 | 62±6*† | api $8.00/M |  | |
| 18 | GPT-5.6 Terra | 62±4*† | api $4.50/M | | |
| 19 | GPT-5.5 Pro | 62±4*† | api $67.50/M | | |
| 20 | GPT-5.4 Pro | 62±4*† | api $67.50/M | | |
| 21 | Claude Opus 5 | 61±4*† | api $10.00/M | | |
| 22 | Claude Fable 5.1 | 60±6*† | api $20.00/M | | |
| 23 | Claude Sonnet 5.5 | 59±6*† | api $4.00/M | | |
| 24 | Claude Fable 5 | 58±4 | api $20.00/M | | |
| 25 | Claude Mythos Preview (limited preview) | 58±4*† | limited preview | | |
| 26 | GPT-5.6 Luna | 57±4*† | api $0.45/M | | |
| 27 | GPT-5.5 | 55±2 | api $11.25/M |  | |
| 28 | Claude Opus 4.6 | 55±6* | api $10.00/M | | |
| 29 | GPT-5.4 | 55±3 | api $5.63/M | | |
| 30 |  Grok 4.20 Multi-Agent (0309) | 55±10* | api $1.56/M | | |
| 31 | Claude Opus 4.8 | 54±2 | api $10.00/M | | |
| 32 | Grok 4.5 | 54±10* | api $3.00/M | | |
| 34 | Claude Haiku 5.5 | 54±6*† | api $0.20/M | | |
| 35 | Claude Sonnet 5 | 53±4 | api $4.00/M | | |
| 36 | Claude Opus 4.7 | 53±5* | api $10.00/M | | |
| 37 | Claude Sonnet 4.6 | 52±3 | api $6.00/M | | |
| 38 | Grok 4.1 Fast (Reasoning) | 52±10* | api $0.28/M | | |
| 39 |  Llama 4 Scout (109B-A17B) | 52±4* | api $0.15/M +local | | |
| 40 | GPT-5.3-Codex | 52±4*† | api $4.81/M | | |
| 41 |  Gemini 3 Flash (Preview) | 52±10* | api $1.13/M | | |
| 42 |  GLM-5.3-Flash | 52±8*† | api $0.24/M | | |
| 43 | Claude Opus 4.5 | 51±10* | api $10.00/M | | |
| 44 |  LongCat 2.0 (1.6T-A48B) | 51±6*† | api $1.30/M | | |
| 45 | GLM-5.2 | 50±3 | api $2.15/M | | |
| 46 | GPT-5.1 | 50±10* | api $3.44/M | | |
| 47 | GPT-5 | 50±10* | api $3.44/M | | |
| 48 | Gemini 3.5 Flash | 50±3* | api $3.38/M | | |
| 49 |  MiniMax-M3 | 49±3* | api $0.48/M | | |
| 50 |  Command A (03-2025) | 49±5* | api $4.38/M +local | | |
| 51 | Claude Sonnet 4.5 | 48±10* | api $6.00/M | | |
| 52 | Muse Glimmer 30B | 48±6*† | api $0.64/M +local | | |
| 54 |  Qwen3.7 Max | 47±3* | api $3.75/M | | |
| 55 | GLM-5.1 | 46±2 | api $2.15/M | | |
| 56 | Qwen3.6 27B | 46±6* | api $1.20/M +local | | |
| 57 | Gemini 3.1 Pro (Preview) | 45±3 | api $4.50/M | | |
| 58 | GPT-5.2 | 45±10* | api $4.81/M | | |
| 61 | GLM-5 | 44±3† | api $1.55/M | | |
| 62 | Llama 4 Maverick (402B-A17B) | 43±6* | api $0.42/M | | |
| 64 | Llama 3.3 70B Instruct | 42±6* | api $0.72/M +local | | |
| 65 |  Kimi K2.6 | 42±3 | api $1.71/M | | |
| 66 | GPT-5 Pro | 42±6*† | api $41.25/M | | |
| 67 | Qwen3.5 122B-A10B | 41±3† | api $1.10/M | | |
| 68 | Gemini 2.5 Pro | 41±10* | api $3.44/M | | |
| 69 | Qwen3.5 397B-A17B | 41±3 | api $1.09/M | | |
| 70 | GPT-5.4 mini | 41±2 | api $1.69/M | | |
| 71 | Qwen3.5 27B | 41±3† | api $0.84/M +local | | |
| 72 | Grok 4.3 | 40±3* | api $1.56/M | | |
| 73 | GPT-4.1 nano | 40±6* | api $0.17/M | | |
| 74 | Qwen3.5 35B-A3B | 39±3† | api $0.35/M +local | | |
| 76 |  Gemma 4 31B | 38±3 | api $0.19/M +local | | |
| 77 | MiniMax-M2.7 | 38±5* | api $0.53/M | | |
| 78 | GLM-4.7 | 38±2† | api $1.00/M | | |
| 79 | GPT-5.4 nano | 38±6*† | api $0.46/M | | |
| 82 |  Nemotron 3 Ultra 550B-A55B | 34±4*† | api $1.35/M | | |
| 83 | Qwen3.6 35B-A3B | 34±4* | api $0.31/M +local | | |
| 84 | GPT-5 mini | 33±6*† | api $0.69/M | | |
| 85 | GLM-4.6 | 31±3† | api $1.00/M | | |
| 86 | Gemini 3.1 Flash-Lite | 31±5* | api $0.56/M | | |
| 88 | Gemma 4 26B-A4B | 23±8*† | api $0.14/M +local | | |
| 89 | GLM-4.5 | 21±4*† | api $1.00/M | | |
| 90 | Claude Haiku 4.5 | 20±6*† | api $2.00/M | | |
| 91 | GLM-4.5-Air | 19±6*† | api $0.42/M +local | | |
| — |  DeepSeek R1 0528 | not yet calibrated | api $0.91/M | | |
| — | DeepSeek V3.1 | not yet calibrated | api $0.42/M | | |
| — | DeepSeek V4 Flash (0731) | not yet calibrated | api $0.11/M | | |
| — | DeepSeek V4 Pro (0813) | not yet calibrated | api $1.63/M | | |
| — | DeepSeek V4.1 Flash | not yet calibrated | api $0.26/M | | |
| — | Gemma 4 12B · Q8_0 | not yet calibrated | local | | |
| — | Gemma 4 E2B · Q8_0 | not yet calibrated | local | | |
| — | Gemma 4 E4B · Q8_0 | not yet calibrated | local | | |
| — | GLM-4.7-Flash | not scored yet | api $0.0000/M +local | ||
| — | gpt-oss-120b | not yet calibrated | api $0.26/M | | |
| — | gpt-oss-20b · mxfp4 | not yet calibrated | local | | |
| — |  Granite 4.2 3B · Q8_0 | not yet calibrated | local | | |
| — |  Hunyuan Hy3 (295B-A21B) | not yet calibrated | api $0.23/M | | |
| — | Hy4 preview (770B-A49B) | not yet calibrated | api $1.25/M | | |
| — |  Inkling (975B-A41B) | not yet calibrated | api $1.73/M | | |
| — | Inkling Small (266B-A12B) | not yet calibrated | api $0.64/M | | |
| — |  K2 Horizon 7B · gguf-q8_0 | not scored yet | local | ||
| — | Kimi K2 Thinking | not yet calibrated | api $1.07/M | | |
| — | Kimi K2.5 | not yet calibrated | api $1.20/M | | |
| — |  LFM2.5 2.6B · q8_0 | not yet calibrated | local | | |
| — | LFM2.5 8B-A1B · q8_0 | not yet calibrated | local | | |
| — |  MiMo-V2.5-Pro (1T-A42B) | not yet calibrated | api $0.65/M | | |
| — |  MiniCPM5 2B · q8_0 | not yet calibrated | local | | |
| — | Nemotron 3 Nano 30B-A3B | not scored yet | api $0.087/M +local | ||
| — | Nemotron 3 Nano 4B · q8_0 | not yet calibrated | local | | |
| — | Nemotron 3 Super 120B-A12B | not scored yet | api $0.28/M | ||
| — |  Ornith 1.0 35B (35B-A3B) · q4_k_m | not yet calibrated | local (slow) | | |
| — | Ornith 1.5 35B-A3B · Q4_K_M | not yet calibrated | local (slow) | | |
| — | Qwen3.5 2B · q8_0 | not yet calibrated | local | | |
| — | Qwen3.5 4B · Q8_0 | not yet calibrated | local | | |
| — | Qwen3.5 9B · Q8_0 | not yet calibrated | local | | |
| — |  Spark-X2.5 1.7B · bf16 | not yet calibrated | local | | |
| — | Spark-X2.5 4B · bf16 | not yet calibrated | local | | |
| — |  Step 3.5 Flash (196B-A11B) | not yet calibrated | api $0.15/M | |
+ 20 more don't fit your machine or budget · show them
+ 1 models with fewer than 3 results left off, since one or two tests say little about overall ability · show them
131 of 135 catalogue models listed here · sources and licences
* new or thinly tested: ranked on too few benchmark results · † only the maker’s own numbers so far · ~RAM is estimated; full context may need more memory