3DGS

ModelJudgedTierSelf-bestRoundsHoursAPI cost
Claude Opus 5.55.82×N15.85×2010.0 h$65.4trace →
Claude Sonnet 5.54.44×N14.06×2010.0 h$56.5trace →
GPT-6.2 Astra3.58×N13.54×5710.0 h$97.4trace →
MiniMax M3.1 Flash2.54×N11.60×2610.0 h$5.47trace →
GPT-6.2 Sol2.35×N12.35×4810.0 h$22.3trace →
Claude Haiku 5.52.35×N12.34×3910.0 h$58.8trace →
GPT-6 Astrasuspected nerf2.04×N12.07×2210.0 h$97.5trace →
GLM-5.31.76×N11.74×2110.0 h$55.2trace →
DeepSeek V4.1 Flash-10171.71×N11.72×529.7 h$4.06trace →
Kimi k3.1-pro1.68×N11.78×3210.0 h$39.6trace →
ModelJudgedSmaller tiles + rewritten binningBypass autograd, activations folded into raster kernelLazy Adam (update only visible Gaussians)Fused loss / SSIMOther main levers
Claude Opus 5.55.82×✓ 8×8, later 8×4; keys generated per slot, tested by bounding-box range; with the new forward / backward kernels 1.23 → 2.01✓ GS_FAST_PIPE=1: whole iteration in one CUDA file, activations and chain rule done in per-Gaussian kernels, no autograd, one sync per step; kernels per step 144 → ~30, 2.01 → 3.02✓ GS_LAZY_ADAM=1, all param groups (positions also made lazy in round 18); 3.42 → 3.68; after folding pending grads into one update and looping 8 rows per thread 4.12 → 4.49SSIM forward's three convolutions merged into one pass, ground-truth mean / variance cached (57 → 38 µs); loss grad fusion tried, no gainbranch-free, barrier-free render and key kernels (render 376 → 208 µs, 4.76 → 5.21), shuffle-free backward, Gaussians reordered in Morton order GS_FAST_SORT=500, depth sort only for visible Gaussians, fused densification, fixed-point atomic adds (training bitwise reproducible)
Claude Sonnet 5.54.44×✓ 8×8 + exact tile enumeration via two-pass warp cooperation (no padding entries in the sort); 8×8 alone nets 0, with enumeration 1.25 → 1.52partial: GS_OPT_RAWRENDER=1 has rasterization take raw params, activations and their backward folded into the kernel (1.52 → 1.86); training loop and raster calls still in PyTorch✓ SH only (48 of 59 floats) GS_OPT_LAZYSH=1, 3.31 → 3.51; position / opacity / scale / rotation stay dense✓ GS_OPT_FASTLOSS=1 loss grad without building an autograd graph (2.56 → 2.62); rewritten SSIM kernel (forward 100 → 44 µs, 3.14 → 3.20)custom single-launch Adam (2.20 → 2.56), SH update on a low-priority stream concurrent with the binning chain (3.81 → 4.06), sort by depth then stable sort by tile (4.06 → 4.24), backward passes only scalar color accumulators (→ 4.35), single-instruction exp / rcp (→ 4.51), fixed-point atomic adds
MiniMax M3.1 Flash2.54×✓ 8×8 alone flat (1.59 vs 1.60); 1.76 with tight bounding boxes; 1.98 with one warp per Gaussian in binning✗ only quaternion normalization made into one bitwise-identical kernel GS_FUSED_NORMALIZE=1 (2.21 → 2.40)✗last commit fused L1 and its grad (2.403 → 2.410, per commit message); SSIM untouchedbitwise-identical fused Adam, desync, backward per-pixel state packed into float4 (inner loop 176 → 129 instructions, 1.43 → 1.62), unused depth path off (→ 2.06), contiguous SH grad writes (→ 2.21)
Claude Haiku 5.52.35×✗ stays 16×16; instead warp-level culling in forward (1.92 → 2.08), binning with one thread per (Gaussian, tile) slot (1.73 → 1.86)✗ only normalization made into one kernel GS_FUSED_NORMALIZE=1 (1.55 → 1.63)✗ dense; rest-SH Adam folded into backward preprocess GS_FUSED_REST_ADAM=1 (1.86 → 1.91)two SSIM kernel attempts (16×16 blocks, redundant flushes removed), both within noise, not committed--use_fast_math (1.35 → 1.43), custom dense Adam, coalesced SH grad writes (1.63 → 1.73), forward color and inverse depth in shared memory (+5.2%), backward visits only possibly contributing pixels, tiles launched in descending length order
GPT-6 Astrasuspected nerf2.04×✓ only in round 20: GS_OPT_TILE_SHAPE=8x8 (keeps the 16×16 parent-tile coverage test, 166.2 → 161.2 s); round 21 compacted sort input GS_OPT_COMPACT_SORT=1 (161.2 → 150.5 s)✗ activations via torch.compile GS_OPT_COMPILED_ACTIVATIONS=1 (234.7 → 225.2 s; of the first two versions one errored, one only 1.08×)✗ dense Adam in Triton (211.9 → 196.4 s)✗ not in the logwarp-cooperative binning for large rectangles (225.2 → 216.0 s), cooperative staging of backward pixel state (187.7 → 170.6 s), background term pre-folded, depth skipped, memory cache kept
GLM-5.31.76×✗ tile size unchanged; last round tightened candidate rectangles GS_OPT_TIGHT_RECT=1 (radix sort −39%); binning one warp per Gaussian (1.55 → 1.58)✗ activations as a pair of standalone kernels outside rasterization GS_OPT_ACT_FUSE=1 (1.60 → 1.70)✗ dense; SH grad and Adam fused into one kernel GS_OPT_SH_ADAM=1 (first version regressed to 1.28, after three reorderings 1.34 → 1.46)only clamp backward fused (−11 µs / step); SSIM only changed to not zero its outputdepth skipped, backward pixel state in shared memory (1.46 → 1.53), four small param groups in one Adam kernel, __expf, densification stats kernel (+0.003)
DeepSeek V4.1 Flash-10171.71×tried, reverted: round 3 switched tiles straight to 8×8, 1.230 vs 1.248 (binning and sort get costlier with tile count); never returned to this directionactivations ✓ folded into preprocess (FUSED_ACT on by default, 1.354 → 1.433); autograd not bypassed✗ log states "state decay needed even for zero grads, the 68% cull rate can't be exploited"✗custom float4 dense Adam (1.26 → 1.35), __expf, forward color staging (kernel −15%), backward per-pixel state as float4 (kernel −16%), two warps per block in backward, bucket boundary state merged into one float4
Kimi k3.1-pro1.68×✗ tile size not in the log; binning only hoisted loop invariants (184.3 → 179.4 s)✗ activation fusion not in the log✗ dense; custom single-kernel Adam only in round 29 (177.8 → 172.5 s)tried, reverted: SSIM scalar backward 0.8 s slower, three revisions crashed within 3000 steps; L1 backward indistinguishable across two timingsdepth skipped (247.5 → 237.8 → 231.3 s), black-background specialization, sample state and pixel state packed into float4, shared forward color, cooperative prefetch of backward state, __expf
↑ Leaderboard