Best value LLM APIs: quality per dollar

Cheap is easy; cheap and good is the real question. This ranks52 chat models by a single derived number — theirLMArena (Chatbot Arena) Elo divided by their cheapest blended price per 1M tokens — so you see how much measured quality each dollar buys. Each model links to its cheapest provider. All figures computed from the index; Elosourced here.

Ranked by value (Elo per dollar)

#ModelEloCheapest /1MVerifiedElo / $
1deepseek-v4-flash · Baidu1436$0.15today9,573
2gemma-4-26b-a4b · Cloudflare1438$0.4today3,595
3gemma-4-31b · CoreWeave1451$0.44today3,298
4glm-4.7-flash · Venice1366$0.46today2,970
5nemotron-3-super-120b-a12b · DeepInfra1361$0.485today2,806
6deepseek-v3.2 · GMICloud1425$0.518today2,749
7minimax-m3 · CoreWeave1443$1.19today1,213
8glm-4.6v · Novita AI1378$1.2today1,148
9minimax-m2.5 · Venice1391$1.22today1,140
10deepseek-v3.1-terminus · AtlasCloud1415$1.25today1,132
11minimax-m2 · Minimax1346$1.275today1,056
12minimax-m2.7 · Novita AI1415$1.35today1,048
13qwen3-next-80b-a3b-thinking · Google1369$1.35today1,014
14minimax-m2.1 · Novita AI1384$1.5today923
15qwen3.5-27b · Alibaba1408$1.755today802
16glm-4.7 · DeepInfra1442$2.15today671
17glm-4.6 · Venice1425$2.18today654
18deepseek-v4 · Alibaba1458$2.3231d ago628
19qwen3.5-122b-a10b · SiliconFlow1417$2.34today606
20glm-5 · GMICloud1458$2.52today579
21kimi-k2.5 · SiliconFlow1451$2.7today537
22qwen3.5-397b-a17b · Alibaba1441$2.73today528
23kimi-k2.6 · Baidu1461$2.856today512
24glm-5.1 · Baidu1466$3.764today389
25qwen3-vl-235b-a22b-thinking · Alibaba1395$4.4today317
26mistral-medium-3-5 · Mistral AI1427$9today159
27claude-fable-5 · Anthropic1507$60today25

Value = Elo ÷ (input + output price per 1M tokens). Since a model's quality is identical across providers, the cheapest host gives the best value — that's the price shown. Models released within the last 12 months; older ones below. Blended is a coarse ranking aid, not a workload estimate; size your own on each model's page.

Older models — released more than 12 months ago (25)

Still priced daily and often still the pragmatic choice, but no longer the current generation.

#ModelEloCheapest /1MAgeElo / $
28gpt-oss-20b · AkashML1317$0.1213 mo10,975
29gpt-oss-120b · AkashML1352$0.213 mo6,760
30gemma-3-27b · DeepInfra1365$0.2417 mo5,688
31qwen3-32b · DeepInfra1347$0.3616 mo3,742
32qwen3-30b-a3b · SiliconFlow1327$0.3916 mo3,403
33llama-4-scout · DeepInfra1321$0.417 mo3,302
34llama-3.3-70b · DeepInfra1317$0.4220 mo3,136
35llama-3.2-3b · Parasail1166$0.3823 mo3,068
36gpt-4o-mini · OpenAI1317$0.7525 mo1,756
37qwen-2.5-72b · DeepInfra1303$0.7623 mo1,714
38command-r · Cohere1226$0.7524 mo1,635
39llama-3.1-70b · DeepInfra1293$0.825 mo1,616
40llama-4-maverick · DigitalOcean1327$0.89617 mo1,481
41glm-4.5-air · Novita AI1373$0.9813 mo1,401
42deepseek-chat-v3 · DeepInfra1396$1.1417 mo1,225
43deepseek-chat-v3.1 · DeepInfra1417$1.212 mo1,181
44qwen3-coder · DeepInfra1387$1.313 mo1,067
45mistral-large · Mistral AI1414$222 mo707
46gpt-4.1-mini · OpenAI1382$216 mo691
47glm-4.5v · Novita AI1353$2.412 mo564
48qwen3-235b-a22b-thinking · Alibaba1400$2.5313 mo553
49deepseek-r1 · DeepInfra1398$2.6515 mo528
50minimax-m1 · Minimax1364$2.614 mo525
51gpt-4o · OpenAI1443$12.527 mo115
52command-a · Cohere1354$12.517 mo108

For contrast: the top of the leaderboard

The highest-quality models by raw Elo — usually not the best value, which is the point.

Elo rankModelEloCheapest /1MElo / $
1claude-fable-51507$6025
2glm-5.11466$3.764389
3kimi-k2.61461$2.856512
4deepseek-v41458$2.323628
5glm-51458$2.52579
6gemma-4-31b1451$0.443,298
7kimi-k2.51451$2.7537
8minimax-m31443$1.191,213
9glm-4.71442$2.15671
10qwen3.5-397b-a17b1441$2.73528

LMArena (Chatbot Arena) · verified 2026-09-05 · leaderboard source · full methodology.

FAQ

What is the best value LLM API right now?

As of 2026-09-05, deepseek-v4-flash leads on quality per dollar: 9,573 Elo points per dollar — 1436 LMArena (Chatbot Arena) Elo at $0.15 per 1M tokens (blended input + output) via Baidu.

How is 'value' calculated here?

Value = a model's LMArena Elo divided by its cheapest blended price per 1M tokens (input + output). A model's quality is identical across providers, so the cheapest host gives the best value — that's the price shown.

Is the highest-Elo model also the best value?

No. The top model by raw Elo is claude-fable-5 at 1507 Elo, but at $60 per 1M its value is only 25 Elo per dollar — well below the value leader deepseek-v4-flash. Top quality rarely means top value, which is the whole point of this ranking.