How much memory does intelligence take? Compare open weights with the models you know.
22 evaluated open-weight models in the catalog · 8 frontier references. The default shows a selection; search explores the full catalog.
9 models in this view
Swipe the chart sideways, or use Table for a compact comparison.
Higher = stronger benchmark result. Left = less estimated memory. Click a point for its source. The Y-axis adapts to the selection.
* USD, US market. Select a machine for the exact configuration, price basis and dated official source. Hardware photos are illustrative. PNG exports the chart and active capacity overlay.
Outside the scatter: DeepSeek V4.1 Flash · max (81.1; memory estimate pending)
Alibaba / Open weightsParameter basis: language component. Dense. Context can be extended up to 1,000,000 tokens with additional configuration; memory requirements change. Technical details checked 2026-09-22.
LiveBench 2026-06-25 · same tasks, 0–100 scale
Q4 weight estimates exclude runtime and cache. Planning memory adds a heuristic allowance; neither is a measured runtime requirement. Category results describe the exact benchmark variant, not a local quantization.
The overall score is the equally weighted mean of seven category means (0–100). All scores use one release. This is not the Artificial Analysis Intelligence Index; the two scales must not be mixed.
CSV ↗ · Categories ↗ · LiveBench ↗ · Data reuse policy ↗Planning formula: total parameters × 0.625 bytes, plus 20% working allowance and 8 GB reserve, rounded up in decimal GB. This is a heuristic, not a measured minimum or guarantee. Quantization, context, KV cache, vision components, runtime and offloading change requirements. MoE uses total parameters, not only active ones.
Qwen counts refer to language parameters; GLM-5.3 and DeepSeek V4 use repository tensor totals; GLM-5.3 Flash uses the manufacturer’s 320B specification. Architectures without a reviewed estimate remain outside the memory axis.
LiveBench evaluates the named model and reasoning effort. Its score is not a measurement of the local Q4 variant. Hardware links are official destinations without affiliate tracking. Stock, exact configuration and regional availability can change.
Existing results are checked when the page is visited and cached for up to 24 hours where supported. No inference is run. If a model disappears, a score shifts over 10 points, or the schema/release changes, the verified saved snapshot stays visible. New releases and hardware mappings require review.