Qwen 3 VL-2B
Qwen 3 VL-2B · Q4_K_M · by Qwen
hf.co/Qwen/Qwen3-VL-2B-Instruct-GGUF:Q4_K_M
Qwen · Q4_K_M · on ollama · vendor build
Measured by one node on hardware we do not own, and signed by each of them. What machines here actually proved — never what the card claims.
matrix built 4 Nov 2026 · every field verified against the signature of the node that produced it, and re-verifiable by you: each signed figure carries the exact bytes the node signed (signing_payload_b64) and the signature over them, and the ed25519 public key is embedded in node_did. effective_ctx is the context length a node PROVED by recall, never the length a model card advertisescontext declared256Kwhat the runtime reported. A node attests it was told this; nobody attests it is true.
context proved by recall4Kthe largest size at which a machine could still find tokens planted across the whole prompt — start, middle and end, all of which had to come back.
Advertises 64× the context any machine here could make it use.Not an accusation — it is what happened here, and it is signed.
Capabilities, measured
Speech outprobed, and failed
The model itself turned text into audio — identified from the container bytes, not from what the engine labelled them, so an error page announced as audio cannot pass. Distinct from this gram running a separate text-to-speech engine: an operator choosing between them is choosing between two architectures.Embeddingsprobed, and failed
The engine was asked for an embedding of a short text and returned a real, non-zero vector.Structured outputpassed
The ENGINE was asked to constrain decoding to a schema, and the answer conformed. This is stronger than politely asking for JSON and hoping: the grammar makes invalid output unreachable, which is what an operator actually builds on.Extended thinkingprobed, and failed
The model emitted reasoning on the engine’s dedicated reasoning channel before answering. Read from that channel rather than by scanning the answer for tags, which is why models that reason are no longer reported as models that do not.Tool callingpassed
The node asked the model to call a function and checked that it called the RIGHT one with a well-formed argument — not merely that it emitted a tool name.Faithful to tool resultspassed
The model was handed a list through a tool and had to report every item back. Calling a tool and understanding its answer are different abilities: a model can produce a perfect call, receive nine items, and confidently report three. This is the flag that separates them.Tool loop finishedpassed
Whether the conversation ENDED after the tool call, or the model kept calling until the turn budget ran out. A model that passes tool calling and fails this will call a tool in production and never come back.Chained tool callsprobed, and failed
The model had to call one tool, read its answer, and call a second with a value derived from it. One call is not agentic work — real specialist jobs are sequences, and a model that never takes the second step looks, from outside, exactly like a specialist silently doing nothing.Counts in imagespassed
Shown a row of shapes and asked how many. Reading rendered text and grasping a scene are different abilities, and a model can do the first with no spatial understanding at all — which is what charts, shelf photographs and document layout actually need.Multiple images at oncepassed
Given two images and asked to compare them. Many engines and models accept one image per turn and silently drop the rest, which turns a comparison into a confident answer about half the evidence.Visionpassed
A four-digit code was rendered into an image and the model had to read it back. A model that cannot see cannot guess four digits, and a model that can has no reason to fail — so this passes and fails for the right reason, which describing a picture never did.Spatial reasoningpassed
Asked which of two shapes is larger and on which side — both ways round, so a coin flip cannot pass. Reading glyphs, counting objects and understanding layout are three different abilities, and only the third answers "what is next to what", which is what charts, diagrams and shelf photographs turn on.Audioprobed, and failed
Given audio and asked to transcribe it.signed by this networkEvery line traces to a node signature.
What machines here proved
context proved by recall4K
peak throughput ever signed9.29 tok/s · did:epn:002408…662a4e
resident while loaded1.9 GiB · did:epn:002408…662a4e
machines that measured it1
samples behind those numbers1
regions it was proved in1
from the public record · unsignedollama /api/show; the node attests that the runtime reported these, not that they are true. Attested by did:epn:002408…662a4e.The only unsigned block on this page. We attest we read these from the runtime — not that they are true. Nobody signs a model card.
What the runtime says about it
context declared256K
parameters1.72B
parameter count1,72,05,74,976
quantisationQ4_K_M
familyqwen3vl
formatgguf
Proven across regions
Measured by
ranked on the provider standingsThe machine that measured it
nothing is averaged — the machine is what you are choosingdid:epn:002408…662a4eGroningen · 9711 · 0.40.14 · on host · 4 Nov 2026Intel Xeon @ 2.20GHz · 2 cores · 31.3 GiB RAM
4Kctx provedtoo slow to answer here at 4K
9.24tok/s now29 samples · 366s1.9 GiB resident · 1.4 GiB on disk
✗ Speech out✗ Embeddings✓ Structured output✗ Extended thinking✓ Tool calling✓ Faithful to tool results✓ Tool loop finished✗ Chained tool calls✓ Counts in images✓ Multiple images at once✓ Vision✓ Spatial reasoning✗ Audio
How far back it remembers, machine by machine
Filled: the needle came back at that prompt size · outlined: asked and lost · blank: not asked · trained for 256K. 0 walks ended because the model forgot, 1 because the machine could not hold the prompt.
| machine · build | 4K | why it stopped |
|---|---|---|
| Gram ·662a4eLinux, CPU only · 4 CPUs · Q4_K_M by Qwen | 4K too slow to answer here |
Carbon per million tokens
Lowest build on each machine · 0 of 1 read off a power sensor, the rest modelled from the parts · fewer grams is better
- Gram ·662a4e307 g5.19 J/tok · modelled
Carbon is location-based: signed energy × the yearly average of the grid where the machine signed its place (EU grid average (applied to NL); Ember Global Electricity Review 2025 (2024 generation), location-based). Facility overhead is not included. A machine on its own solar is not credited until it can prove it.
Run this where your data already is
Tell us the workload, whose machines it should run on, and which jurisdiction it answers to.