The strongest open-weight lines in one view. All three publish weights; the columns show what each costs to run hosted and how they score on the named benchmarks.Last reviewed 2026-09-09.
The GA release of DeepSeek's open 1.6T-parameter MoE flagship for advanced reasoning and long-horizon agents.
The open-weight variant of Qwen3.8 Max, with 95 billion active parameters out of 2.4 trillion total.
NVIDIA's open frontier reasoning model for orchestration, coding, and deep research.
No shared demos for these models yet.
No captured outputs for these models yet.
DeepSeek V4 Pro (0813) has the lowest blended list price. Per 1M tokens: DeepSeek V4 Pro (0813) at $0.66 in / $1.98 out; Qwen3.8 2.4T A95B at $2.00 in / $6.00 out; Nemotron 3 Ultra at $0.60 in / $3.60 out.
DeepSeek V4 Pro (0813) leads with 1M tokens. The other two: Qwen3.8 2.4T A95B at 1M, Nemotron 3 Ultra at 512K.
No, none of the three accepts image input.
This idapt comparison runs all three in aligned columns: the same prompt, plus benchmarks, specs, and prices per model. Open any model in chat from its column to keep working with it.