NVIDIA/cuopt v26.06.00 on GitHub

What's Changed

🚨 Breaking Changes

ci(pr): gate builds on changed-files and broaden safe exclusions by @rgsl888prabhu in #1118
Migrate RMM usage to CCCL MR design by @bdice in #1035
Fold libmps_parser into libcuopt by @mlubin in #1193

🐛 Bug Fixes

Add missing thrust headers across C++ sources by @bdice in #997
Update sonarcube host link to Nvidia by @rgsl888prabhu in #1015
Require openssl so that conda verson is found before system lib by @hlinsen in #1084
fix(cuda_helpers): clear sticky error and avoid cache poisoning in set_shmem_of_kernel by @np96 in #1095
Fix double va_start undefined behavior in error handling by @rgsl888prabhu in #1086
Fixed lower bound in single thread mode by @nguidotti in #1111
test(c_api): skip flaky TimeLimitTestFixture MIP case (#1135) by @rgsl888prabhu in #1136
Replace flaky determinism test by @aliceb-nv in #1141
Use fp64 for all node stack buffers and fix indexing in vehicle order match test by @hlinsen in #1137
fix(benchmarks): drop removed rmm owning_wrapper; build in CI by @rgsl888prabhu in #1143
Fix loose tolerances in the objective function integrality detection by @aliceb-nv in #1148
ci(docker): pin all first-party cuOpt wheels to CUOPT_VER by @rgsl888prabhu in #1157
Fix CI crash on RTX6000 by @nguidotti in #1184
miplib_test: allow tiny error on objective equality by @mlubin in #1199
Fix concurrent LP exception cleanup by @mlubin in #1206
Allow nested parallel regions for determinism mode by @nguidotti in #1201
Fix destruction order in solve.cu by @mlubin in #1216
Fix by-reference capture of local variable in solve.cu by @mlubin in #1214
Fix route priority sort indexing by @bdice in #1222
Drop constness in raft handle for data_model_view_t (routing) by @nguidotti in #1241
Fix lost nodes in branch and bound by @hlinsen in #1138
Fix vehicle fixed cost accounting in fragment-vs-route deltas by @mlubin in #1244
Fix an out of bounds access error by @rg20 in #1245
Fix a bug in dual variables and reduced costs when we add implied bounds on free variables by @chris-maes in #1237
Cuda graph replays on capture error by @akifcorduk in #1253
Fix accidentally quadratic add_slacks_to_basis in crossover by @nguidotti in #1268
Replace constant FJ settings with per-instance device_scalar by @mlubin in #1284
Stabilize logger config in LP BatchSolve and version logging by @mlubin in #1286
Add extra numerical guards for flow cover cuts by @aliceb-nv in #1316
Fix broken sample.mps download link in server LP examples docs by @rgsl888prabhu in #1358
Fix for maximization problems with quadratic constraints by @chris-maes in #1377
Use 26.6 (not 26.06) in README pip install commands by @rgsl888prabhu in #1380

📖 Documentation

Tweak testing instructions in CONTRIBUTING.md by @mlubin in #1000
Exclude L1TESTs from contributing instructions by @mlubin in #1034
CONTRIBUTING.md: Recommend local conda environments by @mlubin in #1085
docs(readme): use py3.13 docker tags (py3.14 not supported) by @rgsl888prabhu in #1112
Fix typo in Constraint docstring by @mlubin in #1113
update mip scaling doc by @Iroy30 in #1121
docs: migrate support link to GitHub Discussions; tidy doc links by @rgsl888prabhu in #1144
Add VRP dimension developer skill by @rg20 in #1233
Fix for cuDSS 0.8; remove incorrect presolve docs in cuopt_cli by @chris-maes in #1353
Update LP/QP/MILP docs to cover 26.06 API additions by @rg20 in #1339
docs: add 26.06 release notes and updates to docs by @rgsl888prabhu in #1294
Mark MILP as beta; add QCQP and SOCP to README by @rgsl888prabhu in #1384

🚀 New Features

Add semi continuous variables support by @hlinsen in #1096
Extend batch pdlp support by @Kh4ster in #1152
Implement python and C api for semi-continuous variables by @hlinsen in #1225
Add LP format reader; accept .lp wherever .mps is accepted by @mlubin in #1120
Detect MIP symmetry using dejavu; Perform orbital fixing and lexical reduction by @chris-maes in #1103
Implement C APIs for adding quadratic objectives and constraints by @rg20 in #1247
Add flow cover cuts at root by @hlinsen in #1178
Extend the barrier solver for Second-Order Cone Programming (SOCP) by @chris-maes in #1290
add QCQP support for grpc by @tmckayus in #1320

🛠️ Improvements

Main merge release/26.04 by @rgsl888prabhu in #998
Update versions1.json as part of version bump by @AyodeAwe in #1004
Main merge release/26.04 1 by @rgsl888prabhu in #1005
Make infra team as default code owner to any files which are not specifically added to a particular group by @rgsl888prabhu in #1013
Bump libcuopt size by 5 mb by @rgsl888prabhu in #1016
Main merge release/26.04 2 by @rgsl888prabhu in #1040
Bump libcuopt wheel size by @rgsl888prabhu in #1087
Replace strtok with thread-safe strtok_r in MPS parser by @rgsl888prabhu in #1088
Update to clang 20.1.8 by @bdice in #1090
bump nightly test to 26.06 by @rgsl888prabhu in #1091
Fix duplicate libmps_parser.so in libcuopt wheel by @rgsl888prabhu in #1092
fix(security): remove insecure SSL verification bypass in dataset downloaders by @rgsl888prabhu in #1108
Latency reductions for concurrent mode by @srib in #1083
Add CTest labels for selective C++ test execution by @rgsl888prabhu in #1102
Improve developer skill with Python bindings, contribution workflow, and agent rules by @rgsl888prabhu in #1106
fix: add retry logic and error handling to dataset download scripts by @rgsl888prabhu in #1116
Update build and test skills by @mlubin in #1127
ci: bump C++ test timeout from 40m to 50m by @rgsl888prabhu in #1130
docs: add entry points for more AI coding agents (Copilot, Cline, Windsurf, Jules, Aider) by @rgsl888prabhu in #1129
Conditional build for remote execution by @nguidotti in #1128
Mark L1 tests as disabled by @mlubin in #1126
Rename .cursor to .agents for codex support by @hlinsen in #1124
docs: replace routing notebook with Python example and drop notebook support by @rgsl888prabhu in #1125
Improve C++ error handling, resource cleanup, and API return checks by @rgsl888prabhu in #1089
ci(images): bump docker CUDA to 13.1.0 and Python to 3.14.1 by @rgsl888prabhu in #1123
ci: assign benchmarks/ to engine codeowners by @rgsl888prabhu in #1142
add a scriptlet to put cuopt_grpc_server on the default path by @tmckayus in #1100
ci: refine CodeRabbit review guide to match cuOpt idioms by @rgsl888prabhu in #1139
build: make docs build opt-in from build.sh by @rgsl888prabhu in #1145
Add QCMATRIX: parsing and quadratic-constraint model by @yuwenchen95 in #1105
Replace custom span with std::span by @mlubin in #1164
skill-evolution: scoped removal and references/ size escape hatch by @rgsl888prabhu in #1160
skill-evolution: add writing style guidance by @rgsl888prabhu in #1159
skill-evolution: drop inline provenance markers by @rgsl888prabhu in #1161
skill-evolution: add thrash-before-landing trigger by @rgsl888prabhu in #1162
Add initial skill evaluation dataset for cuopt-developer by @rgsl888prabhu in #1167
ci(pr): exclude skills//evals/ from test-trigger filters by @rgsl888prabhu in #1170
Add flaky test detection, retry, crash isolation, and nightly reporting by @rgsl888prabhu in #1098
skill-evolution: collapse three-phase model to single workflow by @rgsl888prabhu in #1163
mps_parser: switch ptr+size setters to std::span on mps_data_model_t by @mlubin in #1171
ci(pr): skip build/test jobs for ci/docker-only changes by @rgsl888prabhu in #1158
Add skill evaluation dataset for cuopt-installation-developer by @rgsl888prabhu in #1169
Strip skill-evolution HTML comment markers from SKILL.md files by @rgsl888prabhu in #1174
Add skill evaluation dataset for cuopt-lp-milp-api-python by @rgsl888prabhu in #1172
ci: forward CUOPT_SLACK_MENTION_ID secret to nightly-summary by @rgsl888prabhu in #1182
Unified pseudocost object for the regular and deterministic mode by @nguidotti in #1020
ci(nightly): set fail-fast: false on the build matrix by @rgsl888prabhu in #1186
Combine formulation and API skills into numerical-optimization skill family by @rgsl888prabhu in #1183
ci: format CUOPT_SLACK_MENTION_ID as subteam when ID starts with S by @rgsl888prabhu in #1187
ci(thirdparty-pulp): skip test_numpy_float in nightly cuOpt run by @rgsl888prabhu in #1190
Unified threading model in MIP solver by @nguidotti in #1099
Improve cuopt-developer skill content and sibling-skill routing by @rgsl888prabhu in #1176
Combine installation user skills into cuopt-install by @rgsl888prabhu in #1189
Remove installation restrictions from AGENTS.md by @mlubin in #1200
fix(ci): resolve all zizmor findings and add zizmor pre-commit checks by @gforsyth in #1181
ci: surface crashed-test failures and add per-step summaries (cpp, python, thirdparty) by @rgsl888prabhu in #1191
evals: add eval suite for cuopt-install skill by @rgsl888prabhu in #1205
Run CPUFJ bursts at the root by @aliceb-nv in #1179
Disable flaky swath1 incumbent callback tests by @mlubin in #1208
Reduce memory footprint of cuPDLPx by @Bubullzz in #1153
Minimize Ubuntu Boost dependencies by @bdice in #1165
Require CMake 4.0 by @KyleFromNVIDIA in #1210
fix NodeInfo assert message by @aycsi in #1218
ci: split build.yaml concurrency group by build_type by @rgsl888prabhu in #1211
Build and test with CUDA 13.2.0 by @bdice in #1198
Skip optionally the cost matrix validation check by @hlinsen in #1217
skill: tighten PR-description and CI-authoring guidance for agents by @rgsl888prabhu in #1224
CI: suppress shell:S4830 false positive and scope GHA permissions to job level by @rgsl888prabhu in #1229
sonarqube: auto-detect newest CUDA env file per branch by @rgsl888prabhu in #1236
Revert temporary CUDA 13.2 branch changes by @bdice in #1238
Reduce QP overheads by @rg20 in #1140
skills: add NVIDIA-signed skill cards and signatures (partial set) by @rgsl888prabhu in #1249
Deprecate LP batch solve across Python, server, and docs by @anandhkb in #915
Update lp testing by @Iroy30 in #1110
ci: add NVSkills CI request workflow by @rgsl888prabhu in #1243
Declare defaulted move ops noexcept (SonarQube BLOCKER cpp:S5018) by @rgsl888prabhu in #1227
Dual simplex performance improvements by @chris-maes in #1043
Add new solver setting CUOPT_MIP_PROBING and separate it from presolve by @akifcorduk in #1257
Replaced SpMV with SpMVOp in PDHG for double precision by @Bubullzz in #1226
Add node_limit setting for MIPs by @nguidotti in #1234
ci: fix mangled python image tag on release/26.06 by @rgsl888prabhu in #1269
Make mip_node_t destructor exception-safe (SonarQube CRIT BUG cpp:S1048) by @rgsl888prabhu in #1230
Fix MAJOR SonarQube bugs: mps_parser signed/unsigned compare and example init by @rgsl888prabhu in #1235
Follow-up for LP format by @Iroy30 in #1251
Migrates concurrent root LP to OpenMP by @nguidotti in #1291
build gRPC against OpenSSL 3 by @tmckayus in #1188
io: dlopen libbz2 by versioned SONAME, not unversioned symlink by @rgsl888prabhu in #1297
add mip start attribute by @Iroy30 in #1305
add code generation for grpc boilerplate conversions by @tmckayus in #1107
Detect when MIP objectives must move in discrete steps. Tighten the dual bound using this info. by @chris-maes in #1239
Improve TSP with different vehicle start/end locations by @hlinsen in #1254
skip CuPy 14.1.0 by @jameslamb in #1322
B&B with Local Heaps by @nguidotti in #1149
Add ability to influence the objective gap for SOCP problems by @chris-maes in #1357
Bump libcudss version to 0.8 by @rgsl888prabhu in #1365
Add support for general convex quadratic constraints (x^T Q x + c^T x… by @chris-maes in #1361
Ceil cuDSS to >=0.7,<0.8 by @rgsl888prabhu in #1371

New Contributors

@np96 made their first contribution in #1095
@srib made their first contribution in #1083
@yuwenchen95 made their first contribution in #1105
@Bubullzz made their first contribution in #1153
@aycsi made their first contribution in #1218

Full Changelog: v26.06.00a...v26.06.00