Coding with AI / RTX 5080
RTX 5080
Memory: 16 GB.
Memory bandwidth: 960 GB/s.
Generation speed is bound by the chip's bandwidth; memory decides which models fit.
smartDevstral Small 268% on SWE-bench Verifiedfrozen~58 tok/s · estimated How to run
Devstral Small 2 smart
-
Score: 68% (SWE-bench Verified),
source
-
Speed: 58 tok/s (estimate: bandwidth-bound approximation (scaled from RTX 5060 Ti estimate, 448→960 GB/s))
- Conditions: engine llama.cpp, quant Q4_K_M, context 128
- Min memory: 16 GB
- Command:
llama-server -m devstral-small-2-Q4_K_M.gguf
gpt-oss-20b fast
-
Score: 60.7% (SWE-bench Verified),
source
-
Speed: 204.85 tok/s (measured, source)
- Conditions: engine llama.cpp, quant MXFP4, context 128
- Min memory: 12 GB
- Command:
llama-server -m gpt-oss-20b-MXFP4.gguf
Every number above is either a measurement with a source or a labeled estimate with a method.
Unlabeled estimates do not exist on this site.
Ceiling timeline
The strongest model fitting RTX 5080 (16 GB) at its release date, by SWE-bench Verified.
- 2025-08-05 gpt-oss-20b 60.7%
- 2025-12-09 Devstral Small 2 68%