Every AI model, ranked
We price 432 models across every company hosting them. Of the 123 distinct models that carry an independent quality score, 114 are beaten outright: something else is cheaper and scores at least as well, at a median of 9.2× the necessary cost.
- Models priced
- 432
- Scored on Arena Elo
- 123
- Undominated
- 9
- Host endpoints
- 1,810
The leaderboard
Ranked on quality by default. Filters and workload pricing are off unless you turn them on. The table alone answers most questions.
| Rank | Model | ||||||
|---|---|---|---|---|---|---|---|
| 1 | Anthropic: Claude Opus 5.5 anthropicreasoning billed as output | 1401 | $4.40 | $22.00 | 1.00M | 11 | Claude Opus 5.5 (batch)2.2× cheaper, same quality |
| 2 | Anthropic: Claude Opus 5.5 (batch)batch anthropicreasoning billed as output | 1401 | $2.00 | $10.00 | 1.00M | 11 | value |
| 3 | Anthropic: Claude Sonnet 5.5 anthropicreasoning billed as output | 1382 | $2.00 | $10.00 | 1.00M | 8 | GPT-6.1 Sol (batch)2.0× cheaper, same quality |
| 4 | Anthropic: Claude Sonnet 5.5 (batch)batch anthropicreasoning billed as output | 1382 | $1.00 | $5.00 | 1.00M | 8 | value |
| 5 | OpenAI: GPT-6.1 Sol openaireasoning billed as output | 1372 | $2.20 | $11.00 | 1.05M | 8 | GPT-6.1 Sol (batch)2.2× cheaper, same quality |
| 6 | OpenAI: GPT-6.1 Sol (batch)batch openaireasoning billed as output | 1372 | $1.00 | $5.00 | 1.05M | 8 | value |
| 7 | moonshotaifp4 on some hosts | 1354 | $2.80 | $15.00 | 1.05M | 25 | Muse Spark 1.33.1× cheaper, same quality |
| 8 | MoonshotAI: Kimi K3 (batch)batch moonshotaifp4 on some hosts | 1354 | $2.28 | $11.40 | 1.05M | 25 | Muse Spark 1.32.4× cheaper, same quality |
| 9 | Meta: Muse Spark 1.3 metareasoning billed as output | 1344 | $1.25 | $4.25 | 1.05M | 1 | MiMo-V2.6-Pro4.2× cheaper, same quality |
| 10 | Anthropic: Claude Fable 5.1 anthropicreasoning billed as output | 1344 | $10.00 | $50.00 | 1.00M | 4 | MiMo-V2.6-Pro42× cheaper, same quality |
| 11 | Anthropic: Claude Fable 5.1 (batch)batch anthropicreasoning billed as output | 1344 | $5.00 | $25.00 | 1.00M | 4 | MiMo-V2.6-Pro21× cheaper, same quality |
| 12 | Anthropic: Claude Opus 5 anthropic | 1333 | $5.50 | $27.50 | 1.00M | 11 | MiMo-V2.6-Pro24× cheaper, same quality |
| 13 | Anthropic: Claude Opus 5 (batch)batch anthropic | 1333 | $2.50 | $12.50 | 1.00M | 11 | MiMo-V2.6-Pro11× cheaper, same quality |
| 14 | Xiaomi: MiMo-V2.6-Pro xiaomi | 1328 | $0.435 | $0.870 | 1.05M | 4 | value |
| 15 | Anthropic: Claude Fable 5 anthropicreasoning billed as output | 1317 | $10.00 | $50.00 | 1.00M | 6 | Claude Haiku 5.5 (batch)200× cheaper, same quality |
| 16 | Anthropic: Claude Fable 5 (batch)batch anthropicreasoning billed as output | 1317 | $5.00 | $25.00 | 1.00M | 6 | Claude Haiku 5.5 (batch)100× cheaper, same quality |
| 17 | Meta: Muse Spark 1.2 metareasoning billed as output | 1310 | $1.25 | $4.25 | 1.05M | 1 | Claude Haiku 5.5 (batch)19× cheaper, same quality |
| 18 | Google: Gemini 3.7 Flash googlereasoning billed as output | 1305 | $0.750 | $3.75 | 1.05M | 6 | GLM 5.3 Flash (batch)16× cheaper, same quality |
| 19 | Google: Gemini 3.7 Flash (batch)batch googlereasoning billed as output | 1305 | $0.375 | $1.88 | 1.05M | 6 | GLM 5.3 Flash (batch)7.9× cheaper, same quality |
| 20 | Anthropic: Claude Haiku 5.5 anthropic | 1303 | $0.105 | $0.525 | 1.00M | 10 | GLM 5.3 Flash (batch)2.2× cheaper, same quality |
| 21 | Anthropic: Claude Haiku 5.5 (batch)batch anthropic | 1303 | $0.050 | $0.250 | 1.00M | 10 | tied with GLM 5.3 Flash (batch) |
| 22 | Google: Gemini 3.8 Flash googlereasoning billed as output | 1296 | $0.750 | $3.75 | 1.05M | 6 | GLM 5.3 Flash (batch)16× cheaper, same quality |
| 23 | Google: Gemini 3.8 Flash (batch)batch googlereasoning billed as output | 1296 | $0.375 | $1.88 | 1.05M | 6 | GLM 5.3 Flash (batch)7.9× cheaper, same quality |
| 24 | z-aireasoning billed as outputfp4 on some hosts | 1295 | $1.19 | $4.40 | 1.05M | 41 | GLM 5.3 Flash (batch)20× cheaper, same quality |
| 25 | Z.ai: GLM 5.3 (batch)batch z-aireasoning billed as outputfp4 on some hosts | 1295 | $0.450 | $2.00 | 1.05M | 41 | GLM 5.3 Flash (batch)9.1× cheaper, same quality |
Just launched
Every new release claims to change things. Most arrive already beaten by something cheaper. Here is what actually moved.
15 models launched in the last 60 days. 2 landed on the efficiency frontier. The rest arrived already beaten.
Best model by use case
The overall ranking hides how much the price of good work varies by task. Eight categories are rated head to head in the arena; three more are judged on capability the arena does not measure.
Websites
landing pages and marketing sites
118 models rated head to head in this category. Elo is measured here, not inferred from an overall score.
- 11361 Elo · $1.98 / 1kHighest rated here. Nothing in this category scores above it.Highest rated
- 21312 Elo · $0.472 / 1k4.2× cheaper than the leader, 49 Elo behind.Best value
- 31270 Elo · $0.183 / 1k11× cheaper than the leader, 91 Elo behind.
- 41211 Elo · $0.170 / 1k12× cheaper than the leader, 150 Elo behind.
- 51179 Elo · $0.088 / 1k23× cheaper than the leader, 182 Elo behind.
Priced host by host
For models served by many companies, the same weights carry different prices, context limits and precision. These pages break down every host.
- Morphcheapest$0.055
- Wafer$0.079
- InferenceNet$0.080
- Reka$0.140
- Wafercheapest$0.060
- Wafer$0.190
- Cloudflare$0.190
- InferenceNet$0.198
- OpenInferencecheapest$0.023
- Morph$0.029
- Relace$0.040
- StreamLake$0.043
- OpenInferencecheapest$0.034
- Morph$0.042
- Relace$0.050
- InferenceNet$0.077
- OpenInferencecheapest$0.013
- Relace$0.016
- Sail Research$0.019
- Reka$0.021
- Wafercheapest$0.720
- InferenceNet$0.730
- Sail Research$0.840
- Morph$1.00
- Venicecheapest$0.030
- CoreWeave$0.030
- DekaLLM$0.030
- AkashML$0.030
- Wafercheapest$0.250
- Ionstream$0.370
- Sail Research$0.400
- Sail Research$0.400
- Tenstorrentcheapest$0.040
- Near AI$0.040
- Wafer$0.040
- DekaLLM$0.040
- Inceptroncheapest$0.465
- Chutes$0.500
- DigitalOcean$0.570
- StreamLake$0.598
- OpenInferencecheapest$0.024
- Relace$0.030
- StreamLake$0.083
- DeepInfra$0.090
- Parasailcheapest$0.450
- StreamLake$0.955
- GMICloud$0.957
- DigitalOcean$1.04
Prices are per million input tokens. These are the models hosted by enough companies to make the spread meaningful. The rest of the catalogue is served by one or two hosts, where there is nothing to compare.
What it would cost you
Pick the model you use and the volume you run. The comparison is against the cheapest model of equal quality, so the saving is like for like.
1,200 tokens in · 300 out · 60% cached
$5,314 a year · 50,000 support replies per month
Nothing in the dataset matches this quality for less. This is the efficient choice for the job.
Without changing model at all: Anthropic: Claude Opus 5.5 (batch) is the same weights on a different billing tier at $201.60 a month.
Everything on this page is arithmetic on public data. Nothing is submitted by a vendor and no placement can be bought. Read how cost, quality and best value are worked out, compiled 12 October 2026.