Best Vision LLMs
This page lists models that accept image input alongside text, filtered on catalog modality data. One honesty note up front: the ranking within this list follows general intelligence evidence, because no independent vision-specific benchmark suite is currently ingested into the catalog. General quality is a reasonable but imperfect proxy for visual reasoning, and this page will switch to vision-specific evidence when it is available.
Models marked as served accept OpenAI-style image_url content blocks (including base64 data URIs) through the Inferbase endpoint, so multimodal requests work with unmodified OpenAI SDK code.
The top pick
The strongest measured fit for this task right now, taken from the live ranking below.
- Intelligence Index
- 57.6
- Context
- 1M
The ranking
Ordered from live catalog data; the note beneath the table states exactly how. Missing values are shown as missing, never estimated.
| # | Model | Quality index | Intelligence Index | Context |
|---|---|---|---|---|
| 1 | anthropic | 100 | 57.6 | 1M |
| 2 | anthropic | 99.5 | 56.0 | 1M |
| 3 | anthropic | 99 | 53.4 | 1M |
| 4 | openai | 98.6 | 52.7 | 1.1M |
| 5 | google | 98.1 | 52.6 | 1M |
| 6 | openai | 97.6 | 51.8 | 1.1M |
| 7 | anthropic | 97.1 | 50.8 | 1M |
| 8 | anthropic | 96.7 | 49.6 | 1M |
| 9 | openai | 96.2 | 47.6 | 1.1M |
| 10 | openai | 95.7 | 47.0 | 1.1M |
This list includes models whose catalog specifications declare image input, a published modality claim rather than a verified serving capability. Within that set, ordering follows the general quality index, a 0-100 percentile among the 107 qualifying models with measured benchmark evidence; no vision-specific benchmark suite is ingested yet.
marks models served through the Inferbase API. Missing values are shown as No data rather than estimated.
Also worth evaluating
The next ranks after the top ten, for teams that want a wider shortlist.
Or stop choosing manually
The honest answer changes as benchmarks and prices move. Routing makes the choice per request instead of once.
Send model="auto" to one OpenAI-compatible endpoint and each request is served by the best fit from the same quality and price data behind this ranking, with the decision disclosed per request.
Every routed request records the task it was classified as, the models that qualified, the one that won and why, so a pick made here can be checked against what the router chose in production. A model pinned by name is served as named, with no routing decision.
from openai import OpenAI
client = OpenAI(
base_url="https://api.inferbase.ai/v1",
api_key="YOUR_INFERBASE_KEY",
)
# Let routing pick the best model per request,
# or pin any model id from GET /models.
response = client.chat.completions.create(
model="auto",
messages=[{"role": "user", "content": "..."}],
)Frequently asked questions
How this ranking works and how to act on it.
By general intelligence evidence, clearly labeled as a proxy. Visual reasoning quality correlates with general capability, but they are not identical, and this page states the limitation rather than presenting the proxy as a vision score.
Served models accept OpenAI-style image_url content blocks, either a hosted URL or a base64 data URI, with a detail level knob where the model supports one. Requests are byte-compatible with the OpenAI SDK format.
Nearly all, but not all: Inferbase actively probes its served routes with real image requests, and the rare deployment that rejects images despite a vision-capable base model is surfaced by that verification rather than assumed away.
Providers convert images to input tokens using resolution-dependent formulas, so a high-detail image can cost thousands of input tokens. The per-token prices in this table apply to those converted tokens; batching or downscaling images is the main cost lever.
Yes. When a request contains image content, Inferbase routing considers only image-capable candidates, in the full catalog or within your custom pool, and returns an actionable error if the pool has none, rather than silently dropping the images.
A language model that accepts images alongside text and reasons over both: it can describe a photo, read a chart, extract fields from a scanned form or answer a question about a screenshot. The models here declare image input in their catalog specifications; whether a model also produces images is a separate modality, shown on each model page.
The open-weight entries in this list, and every image-input model in the catalog under the Modalities filter set to Image. Open a model page for its license, its output modalities and its benchmark evidence, and for whether it is served on Inferbase today or listed for reference.
They accept them; how well each reads a dense table or a low-contrast scan varies by model, and no independent vision benchmark is ingested yet to rank that. Test on your own documents before choosing: a served model can be tried with image_url content through the same endpoint your text requests use.
Related rankings
The same evidence-first method, applied to other tasks.
Put intelligence in the middle.
One gateway in front of every model, with your policies applied and every decision on record. Start on the free plan with 5,000 requests a month, no card required.