1.1.18 (2026-10-28)
Features
- bench: discover and benchmark models loaded in LM Studio (#1129) (ccc1390)
- cli: show backend-incompatible models on request (#1143) (eb8db38)
- hardware: detect and show the PCIe link of NVIDIA GPUs (#1153) (b8124fb)
- scraper: backfill release_date for catalog entries without one (#1060) (31a90a4), closes #176
- web: catch the dashboard up with estimate confidence, concurrency, storage and TUI filters (#1094) (363b683)
Bug Fixes
- bench: read OLLAMA_HOST like the Ollama provider does (#1171) (ea1ceb9)
- build: give the Windows binary an 8 MB main-thread stack (#1198) (f2f07d4)
- claim: reject --quant labels it does not recognize (#1162) (628184e)
- claim: size the MoE bandwidth floor from active parameters (#1150) (2491243)
- cli: accept the sizing flags after the subcommand (#1172) (fdc859d)
- cli: exit quietly when stdout is a closed pipe (#1185) (9245299)
- cli: match every word of a multi-word search (#1188) (b7c4812)
- cli: reject --memory and --ram values that do not parse (#1160) (42f2395)
- cli: return structured JSON errors from plan, concurrency and claim (#1191) (a552271)
- diff: compare named models whatever their fit (#1190) (3b4dff2)
- fit: gate pre-quantized FP4 repos on the GPU's kernel support, not the catalog quant (#1127) (781e76a)
- fit: report the quant the MoE offload path chose (#1121) (bb92661), closes #1119
- fit: size GPU-loaded MoE models at the quant they are rated at (#1141) (0490df0)
- fit: size MoE offload for pre-quantized models at their own quant (#1131) (06e236b)
- fit: size usable context from the active experts under MoE offload (#1169) (6ee2785)
- fit: size usable context within the Marginal ceiling (#1126) (a680dd7)
- fit: treat mlx-community repos as Apple-only (#1130) (399b7e9)
- hardware: drop the detected free VRAM when --memory overrides it (#1195) (e293250)
- hardware: keep registry VRAM when PowerShell exits non-zero (#1183) (d47b38b), closes #1180
- hardware: skip legacy Intel iGPUs with lspci controller names (#1148) (bffd1b6)
- mcp: report plan_hardware failures as tool errors (#1193) (d202d9d)
- models: score F32 like F16 instead of a 4-bit quant (#1186) (21bed2c)
- models: size MLA KV cache from the compressed latent (#1145) (85ae435)
- models: stop sanitization hiding sub-260M and decimal-size models (#1104) (3e53c5c)
- plan: estimate F32 speed from 4 bytes per parameter (#1170) (8423800)
- plan: grade paths and upgrade deltas with fit's thresholds (#1125) (4b9fdd9)
- providers: detect llama.cpp installed as single llama binary (#1136) (2cc11f7)
- scraper: keep a rate-limited base lookup out of the params cache (#1071) (29bab4d), closes #1067
- scraper: keep the stated context window for llama3 rope scaling (#1167) (65ddd28)
- serve: accept localhost-only routes over --unix-socket (#1146) (1c0b6bd)
- serve: answer unparsable query strings and bodies with JSON errors (#1192) (176289d)
- serve: only replace a stale socket at the --unix-socket path (#1152) (e915091)
- tui: keep inference bench selection visible (#1177) (687399b)
- tui: replace CPR-dependent Terminal::clear() with same-size fullscreen resize() (#1196) (56d4174)
- update: detect MoE models from config.json expert counts (#1142) (fa80482)
What's Changed
- fix(scraper): keep a rate-limited base lookup out of the params cache by @Akciali in #1071
- feat(scraper): backfill release_date for catalog entries without one by @Akciali in #1060
- fix(fit): report the quant the MoE offload path chose by @jayhemnani9910 in #1121
- fix(models): stop sanitization hiding sub-260M and decimal-size models by @AlexsJones in #1104
- chore(deps): bump signpath/github-action-submit-signing-request from 2 to 3 by @dependabot[bot] in #1076
- fix(fit): size usable context within the Marginal ceiling by @jayhemnani9910 in #1126
- fix(fit): gate pre-quantized FP4 repos on the GPU's kernel support, not the catalog quant by @AlexsJones in #1127
- fix(plan): grade paths and upgrade deltas with fit's thresholds by @jayhemnani9910 in #1125
- fix(fit): treat mlx-community repos as Apple-only by @jayhemnani9910 in #1130
- fix(fit): size MoE offload for pre-quantized models at their own quant by @jayhemnani9910 in #1131
- chore(deps): bump the cargo-minor-patch group with 3 updates by @dependabot[bot] in #1137
- chore(deps): bump jsonschema from 0.55.0 to 0.58.1 by @dependabot[bot] in #1139
- fix(update): detect MoE models from config.json expert counts by @jayhemnani9910 in #1142
- fix(serve): accept localhost-only routes over --unix-socket by @jayhemnani9910 in #1146
- fix(models): size MLA KV cache from the compressed latent by @jayhemnani9910 in #1145
- fix(providers): detect llama.cpp installed as single llama binary by @AstralJaeger in #1136
- Feat/974 tui confidence prefill by @yashchandnani07 in #1003
- [Automated] Weekly benchmark cache refresh by @github-actions[bot] in #1007
- bench: community results for nvidia-geforce-rtx-5060 by @mzambrini in #1080
- bench: community results for apple-m1 by @amino47 in #1081
- chore(deps): bump the github-actions-minor-patch group across 1 directory with 3 updates by @dependabot[bot] in #1092
- feat(web): catch the dashboard up with estimate confidence, concurrency, storage and TUI filters by @sloweyyy in #1094
- feat(bench): discover and benchmark models loaded in LM Studio by @stanleys12 in #1129
- bench: community results for intel-3rd-gen-core-processor-graphics-controller-integrated by @FabioLeitao in #1132
- bench: community results for intel-r-iris-r-xe-graphics by @mrmaskell in #1135
- chore(deps): bump dirs from 6.0.0 to 7.0.0 by @dependabot[bot] in #1138
- [Automated] Weekly model data refresh by @github-actions[bot] in #1140
- fix(fit): size GPU-loaded MoE models at the quant they are rated at by @jayhemnani9910 in #1141
- feat(cli): add --show-incompatible for non-runnable models by @JimmyLiuJia in #1143
- fix(claim): size the MoE bandwidth floor from active parameters by @jayhemnani9910 in #1150
- fix(serve): only replace a stale socket at the --unix-socket path by @jayhemnani9910 in #1152
- fix(cli): reject --memory and --ram values that do not parse by @jayhemnani9910 in #1160
- chore(deps): bump the cargo-minor-patch group with 4 updates by @dependabot[bot] in #1156
- fix(claim): reject --quant labels it does not recognize by @jayhemnani9910 in #1162
- fix(scraper): keep the stated context window for llama3 rope scaling by @jayhemnani9910 in #1167
- fix(hardware): skip legacy Intel iGPUs with lspci controller names by @JimmyLiuJia in #1148
- docs: add WinGet installation instructions for Windows by @Ehsan-hash in #1175
- fix(tui): keep inference bench selection visible by @Ehsan-hash in #1177
- fix(hardware): keep registry VRAM when PowerShell exits non-zero by @Abelo9996 in #1183
- feat(hardware): detect and show the PCIe link of NVIDIA GPUs by @jayhemnani9910 in #1153
- fix(fit): size usable context from the active experts under MoE offload by @spicysharkco in #1169
- fix(plan): estimate F32 speed from 4 bytes per parameter by @spicysharkco in #1170
- fix(bench): read OLLAMA_HOST like the Ollama provider does by @grosjeanbaptiste in #1171
- fix(cli): accept the sizing flags after the subcommand by @grosjeanbaptiste in #1172
- fix(cli): exit quietly when stdout is a closed pipe by @jayhemnani9910 in #1185
- fix(models): score F32 like F16 instead of a 4-bit quant by @spicysharkco in #1186
- fix(cli): match every word of a multi-word search by @spicysharkco in #1188
- chore(deps): bump the cargo-minor-patch group with 6 updates by @dependabot[bot] in #1189
- fix(diff): compare named models whatever their fit by @jayhemnani9910 in #1190
- bench: community results for nvidia-geforce-rtx-5060-ti by @AythaE in #1149
- bench: community results for radeon-rx-7900-xt-7900-xtx by @kortasma in #1151
- bench: community results for radeon-680m by @rdb420 in #1155
- bench: community results for amd-radeon-rx-7900-xt by @rlnrln in #1161
- bench: community results for nvidia-rtx-pro-4000-blackwell-sff-edition by @hide3803 in #1176
- bench: community results for apple-m5-pro by @kruegsche in #1179
- bench: community results for amd-gpu by @FatherlyFox in #1181
- fix(hardware): drop the detected free VRAM when --memory overrides it by @jayhemnani9910 in #1195
- fix(mcp): report plan_hardware failures as tool errors by @jayhemnani9910 in #1193
- fix(serve): answer unparsable query strings and bodies with JSON errors by @jayhemnani9910 in #1192
- fix(cli): return structured JSON errors from plan, concurrency and claim by @jayhemnani9910 in #1191
- fix(build): give the Windows binary an 8 MB main-thread stack by @michaeljabbour in #1198
- test(plan): pin DDR bandwidth in the MoE-offload speed test by @michaeljabbour in #1197
- fix(tui): clear the boot screen without a cursor-position query by @michaeljabbour in #1196
- bench: community results for nvidia-geforce-rtx-3080 by @inkly in #1199
- chore(main): release 1.1.18 by @AlexsJones in #1124
New Contributors
- @AstralJaeger made their first contribution in #1136
- @yashchandnani07 made their first contribution in #1003
- @mzambrini made their first contribution in #1080
- @amino47 made their first contribution in #1081
- @stanleys12 made their first contribution in #1129
- @mrmaskell made their first contribution in #1135
- @JimmyLiuJia made their first contribution in #1143
- @Ehsan-hash made their first contribution in #1175
- @Abelo9996 made their first contribution in #1183
- @spicysharkco made their first contribution in #1169
- @grosjeanbaptiste made their first contribution in #1171
- @AythaE made their first contribution in #1149
- @kortasma made their first contribution in #1151
- @rdb420 made their first contribution in #1155
- @rlnrln made their first contribution in #1161
- @hide3803 made their first contribution in #1176
- @kruegsche made their first contribution in #1179
- @FatherlyFox made their first contribution in #1181
- @inkly made their first contribution in #1199
Full Changelog: v1.1.17...v1.1.18
What's Changed
- fix(scraper): keep a rate-limited base lookup out of the params cache by @Akciali in #1071
- feat(scraper): backfill release_date for catalog entries without one by @Akciali in #1060
- fix(fit): report the quant the MoE offload path chose by @jayhemnani9910 in #1121
- fix(models): stop sanitization hiding sub-260M and decimal-size models by @AlexsJones in #1104
- chore(deps): bump signpath/github-action-submit-signing-request from 2 to 3 by @dependabot[bot] in #1076
- fix(fit): size usable context within the Marginal ceiling by @jayhemnani9910 in #1126
- fix(fit): gate pre-quantized FP4 repos on the GPU's kernel support, not the catalog quant by @AlexsJones in #1127
- fix(plan): grade paths and upgrade deltas with fit's thresholds by @jayhemnani9910 in #1125
- fix(fit): treat mlx-community repos as Apple-only by @jayhemnani9910 in #1130
- fix(fit): size MoE offload for pre-quantized models at their own quant by @jayhemnani9910 in #1131
- chore(deps): bump the cargo-minor-patch group with 3 updates by @dependabot[bot] in #1137
- chore(deps): bump jsonschema from 0.55.0 to 0.58.1 by @dependabot[bot] in #1139
- fix(update): detect MoE models from config.json expert counts by @jayhemnani9910 in #1142
- fix(serve): accept localhost-only routes over --unix-socket by @jayhemnani9910 in #1146
- fix(models): size MLA KV cache from the compressed latent by @jayhemnani9910 in #1145
- fix(providers): detect llama.cpp installed as single llama binary by @AstralJaeger in #1136
- Feat/974 tui confidence prefill by @yashchandnani07 in #1003
- [Automated] Weekly benchmark cache refresh by @github-actions[bot] in #1007
- bench: community results for nvidia-geforce-rtx-5060 by @mzambrini in #1080
- bench: community results for apple-m1 by @amino47 in #1081
- chore(deps): bump the github-actions-minor-patch group across 1 directory with 3 updates by @dependabot[bot] in #1092
- feat(web): catch the dashboard up with estimate confidence, concurrency, storage and TUI filters by @sloweyyy in #1094
- feat(bench): discover and benchmark models loaded in LM Studio by @stanleys12 in #1129
- bench: community results for intel-3rd-gen-core-processor-graphics-controller-integrated by @FabioLeitao in #1132
- bench: community results for intel-r-iris-r-xe-graphics by @mrmaskell in #1135
- chore(deps): bump dirs from 6.0.0 to 7.0.0 by @dependabot[bot] in #1138
- [Automated] Weekly model data refresh by @github-actions[bot] in #1140
- fix(fit): size GPU-loaded MoE models at the quant they are rated at by @jayhemnani9910 in #1141
- feat(cli): add --show-incompatible for non-runnable models by @JimmyLiuJia in #1143
- fix(claim): size the MoE bandwidth floor from active parameters by @jayhemnani9910 in #1150
- fix(serve): only replace a stale socket at the --unix-socket path by @jayhemnani9910 in #1152
- fix(cli): reject --memory and --ram values that do not parse by @jayhemnani9910 in #1160
- chore(deps): bump the cargo-minor-patch group with 4 updates by @dependabot[bot] in #1156
- fix(claim): reject --quant labels it does not recognize by @jayhemnani9910 in #1162
- fix(scraper): keep the stated context window for llama3 rope scaling by @jayhemnani9910 in #1167
- fix(hardware): skip legacy Intel iGPUs with lspci controller names by @JimmyLiuJia in #1148
- docs: add WinGet installation instructions for Windows by @Ehsan-hash in #1175
- fix(tui): keep inference bench selection visible by @Ehsan-hash in #1177
- fix(hardware): keep registry VRAM when PowerShell exits non-zero by @Abelo9996 in #1183
- feat(hardware): detect and show the PCIe link of NVIDIA GPUs by @jayhemnani9910 in #1153
- fix(fit): size usable context from the active experts under MoE offload by @spicysharkco in #1169
- fix(plan): estimate F32 speed from 4 bytes per parameter by @spicysharkco in #1170
- fix(bench): read OLLAMA_HOST like the Ollama provider does by @grosjeanbaptiste in #1171
- fix(cli): accept the sizing flags after the subcommand by @grosjeanbaptiste in #1172
- fix(cli): exit quietly when stdout is a closed pipe by @jayhemnani9910 in #1185
- fix(models): score F32 like F16 instead of a 4-bit quant by @spicysharkco in #1186
- fix(cli): match every word of a multi-word search by @spicysharkco in #1188
- chore(deps): bump the cargo-minor-patch group with 6 updates by @dependabot[bot] in #1189
- fix(diff): compare named models whatever their fit by @jayhemnani9910 in #1190
- bench: community results for nvidia-geforce-rtx-5060-ti by @AythaE in #1149
- bench: community results for radeon-rx-7900-xt-7900-xtx by @kortasma in #1151
- bench: community results for radeon-680m by @rdb420 in #1155
- bench: community results for amd-radeon-rx-7900-xt by @rlnrln in #1161
- bench: community results for nvidia-rtx-pro-4000-blackwell-sff-edition by @hide3803 in #1176
- bench: community results for apple-m5-pro by @kruegsche in #1179
- bench: community results for amd-gpu by @FatherlyFox in #1181
- fix(hardware): drop the detected free VRAM when --memory overrides it by @jayhemnani9910 in #1195
- fix(mcp): report plan_hardware failures as tool errors by @jayhemnani9910 in #1193
- fix(serve): answer unparsable query strings and bodies with JSON errors by @jayhemnani9910 in #1192
- fix(cli): return structured JSON errors from plan, concurrency and claim by @jayhemnani9910 in #1191
- fix(build): give the Windows binary an 8 MB main-thread stack by @michaeljabbour in #1198
- test(plan): pin DDR bandwidth in the MoE-offload speed test by @michaeljabbour in #1197
- fix(tui): clear the boot screen without a cursor-position query by @michaeljabbour in #1196
- bench: community results for nvidia-geforce-rtx-3080 by @inkly in #1199
- chore(main): release 1.1.18 by @AlexsJones in #1124
New Contributors
- @AstralJaeger made their first contribution in #1136
- @yashchandnani07 made their first contribution in #1003
- @mzambrini made their first contribution in #1080
- @amino47 made their first contribution in #1081
- @stanleys12 made their first contribution in #1129
- @mrmaskell made their first contribution in #1135
- @JimmyLiuJia made their first contribution in #1143
- @Ehsan-hash made their first contribution in #1175
- @Abelo9996 made their first contribution in #1183
- @spicysharkco made their first contribution in #1169
- @grosjeanbaptiste made their first contribution in #1171
- @AythaE made their first contribution in #1149
- @kortasma made their first contribution in #1151
- @rdb420 made their first contribution in #1155
- @rlnrln made their first contribution in #1161
- @hide3803 made their first contribution in #1176
- @kruegsche made their first contribution in #1179
- @FatherlyFox made their first contribution in #1181
- @inkly made their first contribution in #1199
Full Changelog: v1.1.17...v1.1.18
What's Changed
- fix(scraper): keep a rate-limited base lookup out of the params cache by @Akciali in #1071
- feat(scraper): backfill release_date for catalog entries without one by @Akciali in #1060
- fix(fit): report the quant the MoE offload path chose by @jayhemnani9910 in #1121
- fix(models): stop sanitization hiding sub-260M and decimal-size models by @AlexsJones in #1104
- chore(deps): bump signpath/github-action-submit-signing-request from 2 to 3 by @dependabot[bot] in #1076
- fix(fit): size usable context within the Marginal ceiling by @jayhemnani9910 in #1126
- fix(fit): gate pre-quantized FP4 repos on the GPU's kernel support, not the catalog quant by @AlexsJones in #1127
- fix(plan): grade paths and upgrade deltas with fit's thresholds by @jayhemnani9910 in #1125
- fix(fit): treat mlx-community repos as Apple-only by @jayhemnani9910 in #1130
- fix(fit): size MoE offload for pre-quantized models at their own quant by @jayhemnani9910 in #1131
- chore(deps): bump the cargo-minor-patch group with 3 updates by @dependabot[bot] in #1137
- chore(deps): bump jsonschema from 0.55.0 to 0.58.1 by @dependabot[bot] in #1139
- fix(update): detect MoE models from config.json expert counts by @jayhemnani9910 in #1142
- fix(serve): accept localhost-only routes over --unix-socket by @jayhemnani9910 in #1146
- fix(models): size MLA KV cache from the compressed latent by @jayhemnani9910 in #1145
- fix(providers): detect llama.cpp installed as single llama binary by @AstralJaeger in #1136
- Feat/974 tui confidence prefill by @yashchandnani07 in #1003
- [Automated] Weekly benchmark cache refresh by @github-actions[bot] in #1007
- bench: community results for nvidia-geforce-rtx-5060 by @mzambrini in #1080
- bench: community results for apple-m1 by @amino47 in #1081
- chore(deps): bump the github-actions-minor-patch group across 1 directory with 3 updates by @dependabot[bot] in #1092
- feat(web): catch the dashboard up with estimate confidence, concurrency, storage and TUI filters by @sloweyyy in #1094
- feat(bench): discover and benchmark models loaded in LM Studio by @stanleys12 in #1129
- bench: community results for intel-3rd-gen-core-processor-graphics-controller-integrated by @FabioLeitao in #1132
- bench: community results for intel-r-iris-r-xe-graphics by @mrmaskell in #1135
- chore(deps): bump dirs from 6.0.0 to 7.0.0 by @dependabot[bot] in #1138
- [Automated] Weekly model data refresh by @github-actions[bot] in #1140
- fix(fit): size GPU-loaded MoE models at the quant they are rated at by @jayhemnani9910 in #1141
- feat(cli): add --show-incompatible for non-runnable models by @JimmyLiuJia in #1143
- fix(claim): size the MoE bandwidth floor from active parameters by @jayhemnani9910 in #1150
- fix(serve): only replace a stale socket at the --unix-socket path by @jayhemnani9910 in #1152
- fix(cli): reject --memory and --ram values that do not parse by @jayhemnani9910 in #1160
- chore(deps): bump the cargo-minor-patch group with 4 updates by @dependabot[bot] in #1156
- fix(claim): reject --quant labels it does not recognize by @jayhemnani9910 in #1162
- fix(scraper): keep the stated context window for llama3 rope scaling by @jayhemnani9910 in #1167
- fix(hardware): skip legacy Intel iGPUs with lspci controller names by @JimmyLiuJia in #1148
- docs: add WinGet installation instructions for Windows by @Ehsan-hash in #1175
- fix(tui): keep inference bench selection visible by @Ehsan-hash in #1177
- fix(hardware): keep registry VRAM when PowerShell exits non-zero by @Abelo9996 in #1183
- feat(hardware): detect and show the PCIe link of NVIDIA GPUs by @jayhemnani9910 in #1153
- fix(fit): size usable context from the active experts under MoE offload by @spicysharkco in #1169
- fix(plan): estimate F32 speed from 4 bytes per parameter by @spicysharkco in #1170
- fix(bench): read OLLAMA_HOST like the Ollama provider does by @grosjeanbaptiste in #1171
- fix(cli): accept the sizing flags after the subcommand by @grosjeanbaptiste in #1172
- fix(cli): exit quietly when stdout is a closed pipe by @jayhemnani9910 in #1185
- fix(models): score F32 like F16 instead of a 4-bit quant by @spicysharkco in #1186
- fix(cli): match every word of a multi-word search by @spicysharkco in #1188
- chore(deps): bump the cargo-minor-patch group with 6 updates by @dependabot[bot] in #1189
- fix(diff): compare named models whatever their fit by @jayhemnani9910 in #1190
- bench: community results for nvidia-geforce-rtx-5060-ti by @AythaE in #1149
- bench: community results for radeon-rx-7900-xt-7900-xtx by @kortasma in #1151
- bench: community results for radeon-680m by @rdb420 in #1155
- bench: community results for amd-radeon-rx-7900-xt by @rlnrln in #1161
- bench: community results for nvidia-rtx-pro-4000-blackwell-sff-edition by @hide3803 in #1176
- bench: community results for apple-m5-pro by @kruegsche in #1179
- bench: community results for amd-gpu by @FatherlyFox in #1181
- fix(hardware): drop the detected free VRAM when --memory overrides it by @jayhemnani9910 in #1195
- fix(mcp): report plan_hardware failures as tool errors by @jayhemnani9910 in #1193
- fix(serve): answer unparsable query strings and bodies with JSON errors by @jayhemnani9910 in #1192
- fix(cli): return structured JSON errors from plan, concurrency and claim by @jayhemnani9910 in #1191
- fix(build): give the Windows binary an 8 MB main-thread stack by @michaeljabbour in #1198
- test(plan): pin DDR bandwidth in the MoE-offload speed test by @michaeljabbour in #1197
- fix(tui): clear the boot screen without a cursor-position query by @michaeljabbour in #1196
- bench: community results for nvidia-geforce-rtx-3080 by @inkly in #1199
- chore(main): release 1.1.18 by @AlexsJones in #1124
New Contributors
- @AstralJaeger made their first contribution in #1136
- @yashchandnani07 made their first contribution in #1003
- @mzambrini made their first contribution in #1080
- @amino47 made their first contribution in #1081
- @stanleys12 made their first contribution in #1129
- @mrmaskell made their first contribution in #1135
- @JimmyLiuJia made their first contribution in #1143
- @Ehsan-hash made their first contribution in #1175
- @Abelo9996 made their first contribution in #1183
- @spicysharkco made their first contribution in #1169
- @grosjeanbaptiste made their first contribution in #1171
- @AythaE made their first contribution in #1149
- @kortasma made their first contribution in #1151
- @rdb420 made their first contribution in #1155
- @rlnrln made their first contribution in #1161
- @hide3803 made their first contribution in #1176
- @kruegsche made their first contribution in #1179
- @FatherlyFox made their first contribution in #1181
- @inkly made their first contribution in #1199
Full Changelog: v1.1.17...v1.1.18