Coding models on Vertex AI, Bedrock and Foundry

LMArena WebDev ranking (overall), published 2026-10-23 · catalogs read 2026-10-25

Top 30

#ModelOrgRatingVertex AIBedrockFoundryOpenRouterIn $/MOut $/M
1Claude Opus 5.5 (Max)anthropic1807Yes · EUYes · EUYes · no EUYes$4.00$20.00
2GPT-6 Astra (Max)openai1774–Yes · no EUYes · EUYes$10.00$50.00
3Claude Sonnet 5.5 (xHigh)anthropic1769Yes · EUYes · EUYes · no EUYes$2.00$10.00
4GPT-6.1 Sol (Max)openai1749–Yes · no EUYes · EUYes$2.00$10.00
5Claude Fable 5.1 (Max)anthropic1742Yes · EUYes · no EUYes · no EUYes$10.00$50.00
6Muse Wonder Full (Max)meta1726––––n/an/a
7Claude Sonnet 5.5 (High)anthropic1709Yes · EUYes · EUYes · no EUYes$2.00$10.00
8Kimi K3.1 Pro (Max)moonshot1692–––Yes$2.00†$10.00†
9Claude Opus 5 (Max)anthropic1691Yes · EUYes · EUYes · no EUYes$5.00$25.00
10GPT-6 Sol (Max)openai1681–Yes · no EUYes · EUYes$2.00$10.00
11Gemini 4 Argon (High)google1674Yes · EU––Yes$2.00$10.00
12Qwen3.8-Max (0902)alibaba1673–––Yes$2.00†$6.00†
13Qwen3.8-Maxalibaba1672–––Yes$2.00†$6.00†
14DeepSeek-V4.1-Flash-20261017 (Max)deepseek1665Yes · no EU–Yes · no EUYes$0.30$1.20
15Claude Opus 5 (High)anthropic1659Yes · EUYes · EUYes · no EUYes$5.00$25.00
16Muse Spark 1.3 (Max)meta1657Yes · no EU––Yes$1.25†$4.25†
17Kimi K3 (Max)moonshot1655–Yes · no EU–Yes$3.00$15.00
18Muse Wonder (Max)meta1651–––Yes$4.00†$20.00†
19Grok 4.7 (xHigh)xai1638Yes · no EUYes · no EUYes · no EUYes$2.00$6.00
20MiMo-V2.6-Flashxiaomi1635–––Yes$0.14†$0.28†
21Hy4 previewtencent1631–––Yes$0.834†$2.501†
22Qwen3.8-Flash-Nextalibaba1629––––n/an/a
23MiMo-V2.6-Proxiaomi1625–––Yes$0.435†$0.87†
24Claude Fable 5 (High)anthropic1624Yes · EUYes · no EUYes · no EUYes$10.00$50.00
25GLM-5.3 (Max)zai1623–Yes · no EU–Yes$1.68$5.28
26Muse Spark 1.3 (xHigh)meta1620Yes · no EU––Yes$1.25†$4.25†
27DeepSeek-V4.1-Flash-20260910 (Max)deepseek1619Yes · no EU–Yes · no EUYes$0.30$1.20
28Grok 4.6 (High)xai1617Yes · no EUYes · no EUYes · no EUYes$2.00$6.00
29GPT-5.6 Sol (xHigh, Codex Harness)openai1615–Yes · no EUYes · EUYes$4.00$20.00
30GLM-5.3-Flashzai1608–––Yes$0.15†$0.50†

Shortlist of cost-optimized frontier models

#ModelOrgRatingVertex AIBedrockFoundryOpenRouterIn $/MOut $/M
3Claude Sonnet 5.5 (xHigh)anthropic1769Yes · EUYes · EUYes · no EUYes$2.00$10.00
4GPT-6.1 Sol (Max)openai1749–Yes · no EUYes · EUYes$2.00$10.00
11Gemini 4 Argon (High)google1674Yes · EU––Yes$2.00$10.00

Open models in the top 30 for up to 256 GB at 8-bit and 4-bit

#ModelLicenseParamsPublished8-bit4-bit
22Qwen3.8-Flash-Nextqwen-community-1.0180BBF16 360 GB180 GB90 GB
30GLM-5.3-FlashMIT321BFP8 328 GB–161 GB

What is what

In the tables

Yes
The provider lists the model in its public catalog, previews included. Quota and access to your account are not covered.
–
Not listed. A row that none of Vertex AI, Bedrock and Foundry lists is grey.
EU
An endpoint that keeps the data in the EU is documented. Global endpoints never count.
no EU
No EU endpoint is confirmed: none on the provider's region pages, the model is missing from them, or two sources disagree. Hover the label for the evidence.
$/M
USD per 1M tokens at the base rate, the lowest enterprise price (from models.dev).
*
The enterprise providers charge different prices.
†
An OpenRouter price, shown only when no enterprise provider has one.
n/a
No price is known.

In the open-models table

Params
Total parameters from the Hugging Face page. A dash for packed weights, whose count cannot be read.
Published
Size of the weight files, with their format. Packed means compressed weights.
8-bit, 4-bit
The weights alone at that many bits per parameter, before any quantization overhead. A dash under 8-bit means it does not fit, or cannot be derived for packed weights; for those the 4-bit figure is the published size, an upper bound (≤).

Where it comes from

Vertex AI
Google Cloud. Its model docs and models.dev; EU from its endpoint locations.
Bedrock
Amazon Web Services. Its model cards and models.dev; EU from an eu.* inference profile, checked against the model card.
Foundry
Microsoft Azure. models.dev; EU from Microsoft's region tables, where Global Standard does not count.
OpenRouter
Personal use, global only. Its public API.
Open models
The leaderboard's license field says which are open. Sizes come from each model's Hugging Face page.