Higher-scoring or cheaper alternatives
139 of the 145 models carrying both an independent score and a standard-delivery price have a higher-scoring or cheaper alternative, with neither dimension worse. Each has a page naming the models and workloads, with recorded capabilities preserved or specific losses identified. The undominated remainder is on the value frontier.
Ordered by the largest measured saving. Prices are blended per million tokens on a balanced workload, as of 2026-10-06. A page here exists only while the finding does — when no candidate improves score or price without worsening the other under a listed workload, the page stops being built.
| Model | Provider | Arena Elo | $/M | Recorded capabilities preserved | Recorded capability losses | Best saving |
|---|---|---|---|---|---|---|
| Claude Opus 4.1 | Anthropic | 1418.6 | $30.00 | 33 | 35 | >99% |
| GPT-4 Turbo | OpenAI | 1271.7 | $15.00 | 83 | 47 | >99% |
| o1 | OpenAI | 1365.8 | $26.25 | 34 | 69 | >99% |
| Claude Opus 4.5 | Anthropic | 1451 | $10.00 | 18 | 11 | −99% |
| Claude Opus 4.8 | Anthropic | 1461 | $10.00 | 8 | 13 | −99% |
| Claude Sonnet 4 | Anthropic | 1339.1 | $6.00 | 37 | 66 | −99% |
| Command R+ (08-2024) | Cohere | 1228.9 | $4.38 | 117 | 7 | −99% |
| GPT-4o (2024-05-13) | OpenAI | 1300.4 | $7.50 | 49 | 71 | −99% |
| GPT-4o (2024-08-06) | OpenAI | 1282.5 | $4.38 | 40 | 73 | −99% |
| GPT-6 Astra | OpenAI | 1442.3 | $20.00 | 5 | 37 | −99% |
| Mistral Large 2407 | Mistral | 1266.1 | $3.00 | 16 | 89 | −99% |
| Claude Sonnet 4.5 | Anthropic | 1438.6 | $6.00 | 15 | 21 | −98% |
| Claude Sonnet 4.6 | Anthropic | 1457.8 | $6.00 | 4 | 14 | −98% |
| GPT-5 | OpenAI | 1406.1 | $3.44 | 8 | 52 | −98% |
| GPT-5.1 | OpenAI | 1422.2 | $3.44 | 6 | 38 | −98% |
| GPT-5.2 | OpenAI | 1412.3 | $4.81 | 11 | 51 | −98% |
| GPT-5.4 | OpenAI | 1452.2 | $5.63 | 0 | 21 | −98% |
| GPT-5.5 | OpenAI | 1466.8 | $11.25 | 0 | 17 | −98% |
| GPT-5.6 Sol | OpenAI | 1456.3 | $8.00 | 0 | 21 | −98% |
| GPT-6 Sol | OpenAI | 1395.4 | $4.00 | 2 | 69 | −98% |
| Claude Sonnet 5 | Anthropic | 1442.9 | $4.00 | 5 | 22 | −97% |
| Gemma 2 27B | 1231.3 | $0.650 | 79 | 0 | −97% | |
| GPT-4.1 | OpenAI | 1383 | $3.50 | 14 | 60 | −97% |
| GPT-5.6 Terra | OpenAI | 1446.3 | $4.50 | 0 | 24 | −97% |
| GPT-6.1 Sol | OpenAI | 1445.6 | $4.00 | 0 | 23 | −97% |
| Grok 4.6 | xAI | 1427.4 | $3.00 | 4 | 37 | −97% |
| Grok 4.7 | xAI | 1399.9 | $3.00 | 5 | 58 | −97% |
| Llama 3.1 70B Instruct | Meta | 1260.9 | $0.720 | 72 | 7 | −97% |
| Mistral Medium 3.5 | Mistral | 1421.2 | $3.00 | 5 | 43 | −97% |
| o3 | OpenAI | 1409.9 | $3.50 | 10 | 49 | −97% |
| o3 Mini | OpenAI | 1319.4 | $1.93 | 12 | 72 | −97% |
| o3 Mini High | OpenAI | 1336.6 | $1.93 | 10 | 69 | −97% |
| o4 Mini | OpenAI | 1353.2 | $1.93 | 9 | 63 | −97% |
| Qwen2.5 Coder 32B Instruct | Qwen | 1230 | $0.745 | 65 | 16 | −97% |
| Claude Haiku 4.5 | Anthropic | 1396.1 | $2.00 | 10 | 44 | −96% |
| Claude Opus 4.7 | Anthropic | 1489.9 | $10.00 | 4 | 2 | −96% |
| Grok 4.5 | xAI | 1448.1 | $3.00 | 3 | 16 | −96% |
| Qwen3.6 Max Preview | Qwen | 1446.8 | $2.31 | 17 | 1 | −96% |
| Claude Sonnet 5.5 | Anthropic | 1466.6 | $4.00 | 3 | 9 | −95% |
| Gemini 2.5 Pro | 1457.8 | $3.44 | 7 | 8 | −95% | |
| GPT-5.4 Mini | OpenAI | 1411.4 | $1.69 | 1 | 43 | −95% |
| Qwen2.5 72B Instruct | Qwen | 1269 | $0.370 | 58 | 2 | −95% |
| Qwen3 Max | Qwen | 1412.4 | $1.56 | 25 | 15 | −95% |
| Qwen3 VL 235B A22B Thinking | Qwen | 1400.7 | $1.30 | 22 | 23 | −95% |
| GLM 5.1 | Z.ai | 1461.5 | $1.83 | 9 | 1 | −94% |
| GLM 5V Turbo | Z.ai | 1437.2 | $1.90 | 7 | 16 | −94% |
| R1 | DeepSeek | 1372.7 | $1.15 | 52 | 5 | −94% |
| GLM 4.5V | Z.ai | 1332.8 | $0.900 | 27 | 35 | −93% |
| Grok 4.3 | xAI | 1396.7 | $1.56 | 0 | 48 | −93% |
| Nova 2 Lite | Amazon | 1361.8 | $0.850 | 3 | 49 | −93% |
| Gemini 3.1 Pro Preview | 1480.2 | $4.50 | 4 | 1 | −92% | |
| Inkling | Thinking Machines | 1441.8 | $1.73 | 0 | 15 | −92% |
| Kimi K3 | Moonshot AI | 1475.9 | $3.76 | 3 | 6 | −92% |
| MiniMax M1 | MiniMax | 1342.8 | $0.850 | 20 | 38 | −92% |
| R1 0528 | DeepSeek | 1427.9 | $0.920 | 22 | 3 | −92% |
| Gemini 2.5 Flash | 1417 | $0.850 | 1 | 29 | −91% | |
| GLM 4.5 | Z.ai | 1430.3 | $1.00 | 17 | 5 | −91% |
| GPT-4.1 Mini | OpenAI | 1340.4 | $0.700 | 5 | 42 | −91% |
| Qwen3 235B A22B | Qwen | 1366.6 | $0.796 | 46 | 5 | −91% |
| Qwen3 235B A22B Thinking 2507 | Qwen | 1415.2 | $0.748 | 16 | 12 | −91% |
| DeepSeek V4 Pro 0423 | DeepSeek | 1445.2 | $1.21 | 3 | 8 | −90% |
| Qwen3.5-122B-A10B | Qwen | 1417.1 | $0.715 | 9 | 17 | −90% |
| Command R (08-2024) | Cohere | 1187.3 | $0.263 | 31 | 4 | −89% |
| DeepSeek V3 | DeepSeek | 1332.5 | $0.500 | 41 | 7 | −89% |
| Gemini 3.5 Flash | 1477.7 | $3.38 | 6 | 1 | −89% | |
| GLM 5 | Z.ai | 1446.3 | $1.09 | 9 | 0 | −89% |
| GPT-5 Mini | OpenAI | 1373.2 | $0.688 | 2 | 34 | −89% |
| Kimi K2.5 | Moonshot AI | 1445.2 | $1.20 | 3 | 8 | −89% |
| MiniMax M2 | MiniMax | 1339.9 | $0.525 | 6 | 35 | −88% |
| MiniMax M2.5 | MiniMax | 1359.3 | $0.521 | 14 | 19 | −88% |
| Qwen3.5 397B A17B | Qwen | 1438 | $0.878 | 4 | 11 | −88% |
| Gemini 3.5 Flash Lite | 1434.7 | $0.850 | 0 | 14 | −87% | |
| Gemini 3.7 Flash | 1487 | $3.00 | 2 | 1 | −87% | |
| GPT-4o-mini (2024-07-18) | OpenAI | 1286.4 | $0.263 | 2 | 25 | −87% |
| Kimi K2.6 | Moonshot AI | 1455.5 | $0.961 | 2 | 5 | −87% |
| Llama 3.3 70B Instruct | Meta | 1274.1 | $0.201 | 24 | 2 | −87% |
| Gemini 3.1 Flash Lite Preview | 1415.7 | $0.563 | 0 | 21 | −86% | |
| Qwen3.5-27B | Qwen | 1408.9 | $0.536 | 8 | 15 | −86% |
| Qwen3.6 Plus | Qwen | 1436.7 | $0.731 | 4 | 8 | −86% |
| Claude Fable 5 | Anthropic | 1491.7 | $20.00 | 4 | 1 | −85% |
| Llama 3.2 3B Instruct | Meta | 1109.5 | $0.120 | 11 | 7 | −85% |
| DeepSeek V3.1 | DeepSeek | 1419.7 | $0.453 | 15 | 2 | −84% |
| DeepSeek V3.1 Terminus | DeepSeek | 1417.1 | $0.453 | 6 | 14 | −84% |
| Inkling Small | Thinking Machines | 1413.7 | $0.638 | 0 | 28 | −84% |
| DeepSeek V3 0324 | DeepSeek | 1375 | $0.438 | 20 | 11 | −83% |
| GLM 4.6 | Z.ai | 1439.9 | $0.760 | 10 | 0 | −83% |
| GPT-5.4 Nano | OpenAI | 1372.1 | $0.463 | 2 | 26 | −83% |
| Mercury 2 | Inception | 1355.2 | $0.375 | 17 | 8 | −83% |
| MiniMax M2.7 | MiniMax | 1404.6 | $0.525 | 5 | 17 | −83% |
| Qwen3.8 27B | Qwen | 1440.7 | $0.562 | 3 | 5 | −83% |
| Solar Pro 4 | Upstage | 1386.2 | $0.525 | 9 | 18 | −83% |
| Trinity Large Thinking | Arcee | 1340 | $0.388 | 11 | 17 | −83% |
| GLM 4.7 | Z.ai | 1435.5 | $0.738 | 8 | 3 | −82% |
| GPT-5.6 Luna | OpenAI | 1431 | $0.450 | 0 | 11 | −82% |
| MiniMax M3 | MiniMax | 1432.1 | $0.485 | 1 | 9 | −82% |
| Qwen3 Next 80B A3B Thinking | Qwen | 1368.3 | $0.413 | 19 | 13 | −82% |
| GPT-4.1 Nano | OpenAI | 1284.8 | $0.175 | 0 | 17 | −80% |
| DeepSeek V3.2 Exp | DeepSeek | 1422.4 | $0.305 | 5 | 9 | −79% |
| Qwen3.5-35B-A3B | Qwen | 1394.4 | $0.355 | 5 | 10 | −79% |
| GLM 4.6V | Z.ai | 1376.8 | $0.450 | 9 | 14 | −78% |
| Muse Spark 1.1 | Meta | 1479.2 | $2.00 | 2 | 2 | −78% |
| Muse Spark 1.2 | Meta | 1483.2 | $2.00 | 1 | 1 | −78% |
| Muse Spark 1.3 | Meta | 1489.9 | $2.00 | 0 | 1 | −78% |
| Qwen3 VL 235B A22B Instruct | Qwen | 1420.1 | $0.370 | 8 | 6 | −77% |
| Qwen3.7 Plus | Qwen | 1454.6 | $0.560 | 4 | 1 | −77% |
| Llama 3.2 1B Instruct | Meta | 1054.6 | $0.071 | 3 | 5 | −76% |
| GLM 4.5 Air | Z.ai | 1383.9 | $0.310 | 9 | 7 | −75% |
| Gemini 3.6 Flash | 1479.5 | $1.50 | 0 | 1 | −74% | |
| Phi 4 | Microsoft | 1216.7 | $0.088 | 13 | 0 | −74% |
| Qwen3 30B A3B | Qwen | 1316.6 | $0.215 | 18 | 5 | −73% |
| Qwen3 Next 80B A3B Instruct | Qwen | 1417.3 | $0.268 | 4 | 7 | −73% |
| DeepSeek V3.2 | DeepSeek | 1424.5 | $0.290 | 9 | 1 | −71% |
| GLM 5.2 | Z.ai | 1470.4 | $1.12 | 2 | 0 | −71% |
| GLM 5.3 | Z.ai | 1471.4 | $1.00 | 0 | 1 | −69% |
| Mistral Small 3 | Mistral | 1233.6 | $0.058 | 2 | 0 | −63% |
| Qwen3 235B A22B Instruct 2507 | Qwen | 1419.3 | $0.205 | 8 | 2 | −62% |
| Claude Fable 5.1 | Anthropic | 1510.6 | $20.00 | 1 | 0 | −60% |
| MiMo-V2.5-Pro | Xiaomi | 1465 | $0.544 | 1 | 1 | −60% |
| Qwen3 32B | Qwen | 1340.1 | $0.130 | 9 | 1 | −59% |
| GPT-6 Luna | OpenAI | 1391.5 | $0.200 | 0 | 9 | −58% |
| GPT-5 Nano | OpenAI | 1320 | $0.138 | 0 | 11 | −57% |
| Gemma 3 27B | 1357.8 | $0.100 | 4 | 3 | −56% | |
| Gemma 4 31B | 1443.4 | $0.205 | 2 | 1 | −56% | |
| GLM 4.7 Flash | Z.ai | 1350.9 | $0.145 | 4 | 5 | −56% |
| Step 3.5 Flash | StepFun | 1403 | $0.150 | 5 | 0 | −53% |
| Qwen3 30B A3B Instruct 2507 | Qwen | 1384.1 | $0.143 | 3 | 4 | −50% |
| Hy3 | Tencent | 1440.7 | $0.231 | 3 | 1 | −46% |
| Granite 4.2 8B | IBM | 1317.8 | $0.108 | 3 | 2 | −40% |
| MiMo-V2.5 | Xiaomi | 1427.5 | $0.175 | 1 | 3 | −38% |
| MiMo-V2.6-Flash | Xiaomi | 1456.4 | $0.175 | 0 | 1 | −38% |
| Qwen3.5-Flash | Qwen | 1397 | $0.114 | 2 | 3 | −36% |
| Gemma 3 12B | 1334.2 | $0.075 | 0 | 2 | −33% | |
| Gemma 3 4B | 1290.8 | $0.063 | 0 | 1 | −30% | |
| Claude Opus 4.6 | Anthropic | 1503.2 | $10.00 | 1 | 0 | −22% |
| Claude Opus 5 | Anthropic | 1502.1 | $10.00 | 2 | 0 | −22% |
| DeepSeek V4 Flash 0423 | DeepSeek | 1432.1 | $0.113 | 1 | 1 | −10% |
| Llama 3.1 8B Instruct | Meta | 1186.5 | $0.025 | 0 | 1 | −9% |
| Gemma 4 26B A4B | 1433.9 | $0.127 | 0 | 2 | −4% | |
| DeepSeek V4.1 Flash | DeepSeek | 1462.5 | $0.182 | 0 | 1 | −3% |
Both columns count candidates with a higher score or a lower price, with neither dimension worse. Recorded capabilities preserved also requires matching or beating the recorded context, maximum output, input modes, tool use and extended reasoning. Recorded capability losses counts candidates with a reduction in at least one of those; each loss is named on the model's page. Savings are rounded to whole percentages, so a saving shown as >99% is not a free model. Scores are LMArena Elo, used under CC BY 4.0 from the official dataset. Models with no independent score are absent from this comparison in both directions — unrated is not a zero.
Questions this page answers
What does dominated mean on this index?
At least one other standard-delivery catalogue model has a higher score or a lower price, with neither dimension worse, under a listed workload. The columns distinguish preserved recorded capabilities from named losses. Models with no independent score are absent in both directions.
What distinguishes the two kinds of alternative?
Both improve score or price without worsening the other. Recorded capabilities preserved means matching or beating the recorded context, maximum output, input modes, tools and extended reasoning. Recorded capability losses means at least one of those is reduced; each loss is named on the model page.
Why can a page disappear?
Each /alternatives/<slug>/ page is built only while a candidate improves score or price without worsening the other under a listed workload. When that comparison no longer holds, the page stops being generated.