FlagOS Daily Intelligence Report (2026-08-21)
Research window: 2026-08-20 10:18 ~ 2026-08-21 10:18 Beijing Time Sources: GitHub (org: flagos-ai), Google News aggregation (Chinese and English), HN, BAAI Community, vendor official sites, etc. (see appendix for details)
Index
- I. Open-Source Project Progress (GitHub Activity)
- 1.1 FlagGems leads with 27 commits: batch KernelGen operator onboarding, Moore Threads SQMMA matrix multiplication optimization, multi-chip fixes (08-20)
- 1.2 build-infra logs Hygon vllm 0.24.0 dual-compiler E2E, app image deferred pending FlagTree #1020 (08-20)
- 1.3 FlagBLAS completes triangular solver family TPSV/TRSV/TBSV, Ascend CI merged (08-20~08-21)
- 1.4 FlagSparse completes CUDA and DCU code merge (08-20)
- 1.5 FlagPrism adds Iluvatar CoreX Tianshu/CoreX debugging support (08-20)
- 1.6 Other repository activity: FlagTree / FlagCX / FlagTensor / flir / FlagGems-vllm (08-20~08-21)
- II. News Coverage and Ecosystem
- 2.1 Overall assessment of news coverage
- 2.2 Ecosystem reference items: Moore Threads and Hygon H1 financial reports, Cambricon share price breaks 1,000 yuan after interim report, MetaX distribution system (08-20)
- 2.3 Trends over the past 3 days
- III. Member Company Deep Dive
- IV. Summary
- Appendix: Complete Source List
1. Open-Source Project Progress (GitHub Activity)
Window Overview: Of the 52 repos in the org, 14 had pushes during the window; commit search returned 55 in-window commits (FlagGems 27, FlagBLAS 12, leading the pack), with no formal GitHub Release. Two headlines this window: high-density expansion of the FlagGems operator matrix (KernelGen batch operators + Moore Threads SQMMA deep optimization) and build-infra completing dual-compiler E2E validation records for Hygon vllm 0.24.0; additionally, FlagSparse merged its CUDA/DCU backends, FlagPrism added Iluvatar debugging support, and FlagBLAS completed its triangular solver family.
1.1 FlagGems: 27 commits lead the pack, KernelGen batch operators + Moore Threads SQMMA matmul optimization (window headline)
Source: FlagGems commits
- KernelGen Nvidia operator batch landing (8/20 10:33-10:53, 12 operators): special_expit (#4263), linalg_householder_product (#5179), adaptive_max_pool2d (#5077), rshift (#5075), amp_update_scale (#5503), _list_to_tensor (#5273), fake_quantize_per_channel_affine_cachemask (#5455), _convolution_double_backward (#5505), view_as_complex (zero-copy view, #5373), _batch_norm_impl_index_backward (#5504), adaptive_avg_pool1d (#5550), float_power (#5517).
- KMCompiler multi-chip expansion and fixes (16:23-17:50): linalg_matrix_norm (Nvidia #5595, Iluvatar/Hygon #5648), dist operators (Nvidia/MetaX #5406), masked_scatter_backend (#4954), index_fill/index_fill_ (Nvidia/Ascend #5643), Ascend tle import fix (#5616) and bug fixes (#5641/#5654), Iluvatar bug fix (#5644), fp64 benchmark exclusion (iluvatar/ascend, #5658).
- MTHREADS (Moore Threads) matmul deep optimization: addmm_out optimized with SQMMA/FMA override (#5604, 18:19), baddbmm optimized with SQMMA (#5666, 19:42).
- Other: Enflame 1.10.6 backend flagtree version update (#5649, 14:04), Ascend erfc fallback implementation (#5380, 14:20), T-Head PPU upsample/quant operator test fixes (#5655, 15:36).
Interpretation: Echoing the 8/18 “KernelGen batch shipping across all vendors,” this window saw the KernelGen pipeline land 12 operators in a single day on the Nvidia side, while KMCompiler saw cross-chip (Nvidia/Ascend/Hygon/Iluvatar/MetaX) fixes and expansion, with operator coverage continuing to converge toward PyTorch high-level and quantization/backward APIs; the SQMMA hardware-instruction-level optimization of addmm_out/baddbmm on the Moore Threads side marks the operator library’s shift from “coverage” to “single-chip deep tuning.”
1.2 build-infra: Hygon vllm 0.24.0 dual-compiler E2E validation records (#466), app image deferred pending FlagTree #1020
Source: PR #466, PR #464, PR #465
- #466 (merged 8/20 20:19): Records dual-compiler validation of hygon (Hygon, DTK 26.04) vllm 0.24.0 on hygon25 nodes — both the F path (flagtree 0.6.1+hcu3.6, temporarily bypassed using the container’s default cluster_dims value, with FlagTree PR #1020 as the upstream fix) and the T path (vendor triton 3.5.1) reached Application startup complete with consistent inference; hygon app image deferred: the F-path default compiler is blocked by FlagTree #1020, and the packaging/flagtree/hygon wheel needs rebuilding alongside a new flagtree release; the vllm validation matrix hygon 0.24.0 T/F cells turned green.
- #464 (8/20 11:00): megatron-training app image build records + h20 dual-compiler validation.
- #465 (8/20 12:21): sunrise 0.20.2 app-image handoff documentation.
- FlagTree #1020 (created 8/20 17:23, pending merge): default cluster_dims=(1,1,1) for backends that omit it — fixes the AttributeError during vllm 0.24.0 KV-cache warmup on backends without cluster-launch support such as hcu/mthreads (reproduced at runtime on hygon DTK 26.04).
Interpretation: Following yesterday’s full-pipeline delivery for Sunrise, Hygon becomes the next inference backend deeply validated by build-infra. The cluster_dims fix that the Hygon F path (FlagTree compiler) depends on has been filed as #1020; once merged, the hygon flagtree wheel will be rebuilt and the app image refreshed — the “validation records first, release mechanism follows” cadence matches the SUNRISE path.
1.3 FlagBLAS: Triangular solver family completed, Ascend CI merged (12 commits)
Source: FlagBLAS commits
- Level-2 triangular solver family (AriesMTR series PRs #82/#83/#85/#87, 8/20 13:33-16:37): TPSV (triangular banded solve), TRSV (triangular solve), and TBSV support merged in succession, with benchmarks updated in sync.
- NVIDIA CTPMV row-major parameter handling fix (10:24); removed the Ascend axpy demo operator (16:52).
- Engineering governance: pyproject.toml fix (#89, 21:03); Ascend CI merged (bin913/ascend_ci, 8/21 10:01 Beijing time).
Interpretation: Continuing yesterday’s L2 operator family cadence, the completion of the triangular solvers (TRSV family) advances FlagBLAS Level-2 coverage another notch; the Ascend CI merge indicates the BLAS library has also entered the multi-chip CI matrix stage.
1.4 FlagSparse: CUDA and DCU code merged (5 commits)
Source: FlagSparse commits
8/20 20:51-21:41: “cuda and dcu merge” — unified merge of CUDA and Hygon DCU backend code paths (#42/#44, including the NCIC-AlphaSparse upstream merge), with CI refinement advancing in parallel. The sparse computing library is beginning to converge and unify at the backend implementation level, reducing dual-path maintenance costs.
1.5 FlagPrism: Added Iluvatar Tianshu/CoreX debugging and profiler support (#2)
Source: PR #2 (merged 8/20 18:43)
FlagPrism Debugger adds Tianshu/CoreX backend support: Triton kernel debugging and summary instrumentation capabilities extended to this hardware platform; backend name resolution now includes tianshu/corex/iluvatar, with build options, documentation, and CoreX/Tianshu example scripts updated in sync. FlagPrism’s chip coverage continues to expand with its debugger backend matrix.
1.6 Other Repository Activity
- FlagTree (1): [Metax][TLE] Metax TLE support local pointer (#971, 17:17) — MetaX TLE local pointer support.
- FlagCX (2): [CRL] homogeneous communicator uses private vendor unique ID (#550, 14:29); action-download-artifact dependency upgrade (#547, 11:34).
- FlagTensor (1): merged iluvatar branch (#13, 10:22) — Iluvatar backend support landed.
- flir (1): [TLE-Distributed] distributed remote_ptr unstructured load offset analysis (#66, 19:27).
- FlagGems-vllm (2): absolute import (#700, 8/21 09:38), CODEOWNERS update (#701, 09:45).
- Branch pushes (non-default branches, not merged to main): docs (modelscope auto-sync branch 20260820), whisper-triton (dependabot), FlagTree-AscendNPU-IR (tle-dev and other dev branches), vllm-plugin-FL (v0.3.0-dev and other dev branch groups) — pushed_at within the window but no new commits on main.
- Quiet repositories: FlagScale, KernelGen, FlagPerf, FlagAttention, FlagQuantum, FlagOS-Robo, Megatron-LM-FL (main), sglang-plugin-FL, Torch-FL, and others had no new commits during the window.
II. News Coverage and Ecosystem
2.1 Overall Assessment of News Coverage
Google News returned zero direct hits within the window for FlagOS/FlagGems/FlagScale/FlagTree/FlagPerf/FlagCX/KernelGen component keywords (consistent with 8/20); HN yielded only 3 irrelevant hits (HN ban discussion, Stripe alternatives, macOS editors — all excluded). The “BAAI open source” query returned mostly generic AI articles (Kimi K3 reviews, etc., with no substantive connection to FlagOS); queries for Enflame/Iluvatar CoreX/Tsingmicro were entirely noise hits (Hongxin Electronics interim report, Biren Technology quote page, Hwatsing equipment shipment, entertainment/livelihood items — all excluded as a batch). Component-level news remains quiet, with valid items concentrated on member organizations’ commercial developments and ecosystem partnerships (see 2.2 and Chapter III for details).
2.2 Ecosystem Reference Items (within window)
| Date | Event | Source |
|---|---|---|
| 8/20 | Moore Threads H1 revenue RMB 1.736 billion, up 147% YoY, losses sharply narrowed but operating cash flow negative (H1 financial report); follow-up commentary “Hong Kong IPO: Domestic GPU Crown Recalibrated” | Huxiu, Caijinghao |
| 8/20 | Hygon H1: AI Agents drive revenue surge of over 60%, but cash flow turns negative; continues to deepen “chip + ecosystem” development strategy | Sohu, Eastmoney |
| 8/20 | Hygon connects the full chain from film production to broadcast, domestic substitution in broadcasting and audiovisual enters deep-water zone | Sina Finance, CSDN |
| 8/20 | Hygon empowers Hefei token economy (“from able to compute to computing well”); Esri partners with Hygon to build self-controllable GIS ecosystem; hosts research visit from 20 institutions including BOCOM Schroders | Sina Finance, Sohu, Eastmoney |
| 8/20 | Cambricon H1 net profit RMB 2.311 billion, up 122% YoY; share price fell below RMB 1,000 on first trading day after interim report, market cap retraced nearly 40% | Guandian, Sohu |
| 8/20 | Xingyun Technology subsidiary obtains MetaX “Industry Ark” mainland China master distribution rights; in-depth report on “Yunjian MetaX” space-based intelligent computing industry logic (follow-up on 7/24 launch event) | Sina Finance, Sina Finance |
| 8/20 | BAAI Community publishes “Meet AI Compiler Episode 9” event recap (8/1 Beijing salon): Tencent FalconGEMM low-complexity matrix multiplication, Huawei Bisheng compiler component AscendNPU IR fully open-sourced (supporting Triton and other operator ecosystems), general-purpose compiler sharing for embodied intelligence | BAAI Community hub.baai.ac.cn |
Excluded items: Securities Star/Eastmoney ETF share reports and main capital flow (heavy holdings of Cambricon, MetaX, Moore Threads, daily routine reports); Cambricon margin trading balance data; generic industry roundups such as “Full Analysis of 20 Domestic Semiconductor IPOs”; Unitree Technology IPO visit (non-member organization); Biren Technology quote page; Hongxin Electronics interim report; entertainment/livelihood noise.
2.3 Recent 3-Day Trends (corroborated by GitHub org pushed)
| Date | Active Repos | Main Content |
|---|---|---|
| 8/19 | build-infra(24), FlagBLAS(19), Torch-FL(6), FlagScale(3), etc. — 14 repos | SUNRISE wheel pipeline launch + vllm 0.24.0 E2E; Kunlunxin joins vllm plugin line; FlagBLAS L2 operator family completion; Torch-FL three-chip AMP/profiler |
| 8/20 | FlagGems(27), FlagBLAS(12), FlagSparse(5), build-infra(3), etc. — 12 repos | KernelGen Nvidia 12 operators in one day + KMCompiler multi-chip fix expansion; hygon 0.24.0 dual-compiler E2E records; FlagSparse CUDA/DCU merge; FlagBLAS triangular solver family; FlagPrism Iluvatar CoreX support |
| 8/21 (end of this window) | FlagGems-vllm(2), FlagBLAS(1) | FlagGems-vllm engineering maintenance (absolute import/CODEOWNERS); FlagBLAS Ascend CI merged; hygon app image postponement decision (pending #1020) landed |
Trend assessment: Following 8/18’s “KernelGen full-vendor batch shipment,” this window shows dual mainlines of “operator library density expansion + inference backend validation deepening.” On the operator side, FlagGems’ 27 commits in a single repo (KernelGen pipeline batch operators, KMCompiler multi-chip matrix, Moore Threads SQMMA deep tuning) indicate the unified operator library has entered a phase of “broad coverage + per-chip fine-tuning” in parallel; on the inference side, build-infra has launched Hygon hygon 0.24.0 dual-compiler E2E validation (F/T dual paths) after SUNRISE, with validation records and release mechanisms (app image, wheel rebuild) advancing in sync, making Hygon the next deeply integrated inference backend. It has been over 8 weeks since FlagOS 2.1 (6/24), and this window still shows no formal release signal, but hygon flagtree wheel rebuild and app image refresh will be near-term observation points.
III. Deep Dive on Member Organizations
Multi-vendor activity map within the window:
| Vendor | Chip/Backend | Activity within window | Evidence |
|---|---|---|---|
| BAAI | — | build-infra led hygon 0.24.0 E2E record (#466) and SUNRISE 0.20.2 handoff (#465); BAAI community published Meet AI Compiler #9 recap | build-infra commits / hub.baai.ac.cn |
| Hygon | Hygon DCU | build-infra #466: vllm 0.24.0 dual-compiler (F/T paths) E2E verification passed, app image deferred; FlagSparse cuda/dcu backend merged; FlagTree #1020 (KV-cache warmup fix, pending merge); FlagGems Ascend/Hygon fixes (#5654) and linalg_matrix_norm (#5648); Commercial: H1 revenue up over 60%, film/TV production, GIS, and Hefei token economy ecosystem deployments | build-infra #466 / FlagSparse / FlagTree #1020 / FlagGems / multiple news sources |
| Moore Threads | MUSA | FlagGems addmm_out SQMMA/FMA (#5604) and baddbmm SQMMA (#5666) optimizations; Commercial: H1 revenue RMB 1.736B +147%, losses sharply narrowed, operating cash flow negative, Hong Kong IPO commentary follows | FlagGems #5604/#5666 / Huxiu / Caijing |
| Iluvatar CoreX | Iluvatar/CoreX | FlagPrism Tianshu/CoreX profiler+debugger support (#2); FlagTensor iluvatar branch merged (#13); FlagGems Iluvatar fixes (#5644) and linalg_matrix_norm (#5648) — multi-track progress within the window | FlagPrism #2 / FlagTensor #13 / FlagGems |
| MetaX | MetaX | FlagGems dist operator support (#5406, Nvidia/MetaX); FlagTree Metax TLE local pointer (#971); Commercial: Xingyun Technology obtains “Industry Ark” general distribution rights | FlagGems #5406 / FlagTree #971 / Sina Finance |
| SUNRISE | SUNRISE | build-infra sunrise 0.20.2 app-image handoff document (#465) — consolidation of yesterday’s full-pipeline delivery | build-infra #465 |
| Enflame | Enflame GCU | FlagGems enflame 1.10.6 backend flagtree version update (#5649) | FlagGems #5649 |
| Ascend (outside ecosystem) | Ascend | FlagGems erfc rollback (#5380), tle import and bug fixes (#5616/#5641); FlagBLAS Ascend CI merged (bin913/ascend_ci) | FlagGems / FlagBLAS |
| T-Head | Zhenwu PPU | FlagGems PPU upsample/quant test fixes (#5655) | FlagGems #5655 |
| Cambricon | Cambricon MLU | No new open-source activity; Commercial: H1 net profit RMB 2.311B +122%, share price fell below RMB 1,000 after interim report | Viewpoint / Sohu |
| Tsingmicro / Horizon Robotics / Kunlunxin | TXDA / BPU / KunlunXin | No open-source or news activity within the window (Kunlunxin entered the vllm plugin track yesterday; Horizon Robotics had Journey 6B design-win news on 8/19) | — |
Trend assessment: The keyword for the chip landscape in this window is “parallel breadth in operators and depth in inference.” On the inference side, Hygon has succeeded SUNRISE as the deep-verification target for build-infra (hygon25 node, DTK 26.04, F/T dual compilers, app image decision), with FlagTree #1020 being the critical blocker for formalizing Hygon’s inference pipeline; on the operator side, Moore Threads (SQMMA matrix multiply), Iluvatar CoreX (FlagPrism debugger + FlagTensor merge + FlagGems fixes), and MetaX (dist operators + TLE) all made substantive progress, with member participation shifting from “onboarding” to “optimization.” On the commercial front, Moore Threads’ H1 revenue surge, Hygon’s H1 growth and multi-industry deployments, and Cambricon’s interim net profit doubling are the highlights; Cambricon’s share price falling below RMB 1,000 and related capital flow fluctuations are routine market movements, noted for reference only.
IV. Summary
- FlagGems leads with 27 commits in a single repo (one of the window’s headlines): KernelGen Nvidia batch-landed 12 operators in a single day, KMCompiler cross-chip (Ascend/Hygon/Iluvatar CoreX/MetaX) fixes and extensions, and Moore Threads addmm_out/baddbmm SQMMA matrix multiply deep optimization — the operator library has entered a phase of “breadth coverage + per-chip fine-tuning” in parallel.
- Hygon hygon inference pipeline verification deepens (the window’s second headline): build-infra #466 records vllm 0.24.0 dual-compiler E2E (hygon25, DTK 26.04; F path flagtree 0.6.1+hcu3.6, T path vendor triton 3.5.1 both reaching startup completion); app image deferred pending FlagTree #1020 (cluster_dims default value fix, submitted 8/20) merge and hygon wheel rebuild.
- Sparse and debugging toolchain expansion: FlagSparse completed CUDA/DCU backend code merge; FlagPrism Debugger added Iluvatar CoreX Tianshu/CoreX support; FlagTensor merged the iluvatar branch — Iluvatar CoreX advanced on multiple tracks in this window.
- FlagBLAS triangular solver family completed: TPSV/TRSV/TBSV merged in succession (AriesMTR series PR #82-#87), Level-2 coverage advances another notch; Ascend CI merged, multi-chip CI matrix extended to the BLAS library.
- Member organizations active on the commercial front: Moore Threads H1 revenue RMB 1.736B +147%; Hygon H1 revenue up over 60% with film/TV production, GIS, and Hefei token economy ecosystem deployments; Cambricon H1 net profit RMB 2.311B +122%; MetaX “Industry Ark” distribution system expanded.
- Quiet on the news side: zero hits on component keywords, no official GitHub Release; Hygon flagtree wheel rebuild and app image refresh are near-term release signal observation points.
Limitations: Commit counts and merge times come from commit search (the search index may lag slightly behind actual merges; individual commit times are based on repos API pushed_at); the FlagBLAS series commits are primarily batch merges from AriesMTR contributors; gnews redirect links cannot be scraped for full text, so news-side judgments rely on headlines and multi-source cross-verification; Hygon/Moore Threads/Cambricon H1 financial data is based on media reports and has not been verified item-by-item against original company announcements.
Appendix: Complete Source List
| No. | Event | Source Link |
|---|---|---|
| 1 | FlagGems KernelGen Nvidia 12 operators (#4263/#5179/#5077/#5075/#5503/#5273/#5455/#5505/#5373/#5504/#5550/#5517) | https://github.com/flagos-ai/FlagGems/commits/main |
| 2 | FlagGems KMCompiler multi-chip expansion (#5595/#5406/#4954/#5643/#5648/#5616/#5641/#5654/#5644/#5658) | https://github.com/flagos-ai/FlagGems/commits/main |
| 3 | FlagGems MTHREADS SQMMA (#5604/#5666), enflame bump (#5649), erfc (#5380), PPU fix (#5655) | https://github.com/flagos-ai/FlagGems/commits/main |
| 4 | build-infra hygon 0.24.0 dual-compiler E2E record (#466) | https://github.com/flagos-ai/build-infra/pull/466 |
| 5 | build-infra megatron app image (#464), sunrise 0.20.2 handoff (#465) | https://github.com/flagos-ai/build-infra/commits/main |
| 6 | FlagTree #1020 cluster_dims fix (pending merge) | https://github.com/flagos-ai/FlagTree/pull/1020 |
| 7 | FlagBLAS triangular solver family (#82/#83/#85/#87), CTPMV fix, Ascend CI | https://github.com/flagos-ai/FlagBLAS/commits/main |
| 8 | FlagSparse cuda/dcu merge (#42/#44) | https://github.com/flagos-ai/FlagSparse/commits/main |
| 9 | FlagPrism Tianshu/CoreX support (#2) | https://github.com/flagos-ai/FlagPrism/pull/2 |
| 10 | FlagTree Metax TLE (#971); FlagCX CRL (#550); FlagTensor iluvatar (#13); flir #66; FlagGems-vllm #700/#701 | https://github.com/flagos-ai/FlagTree/commits/main |
| 11 | org repos overview (52 repos) | https://api.github.com/orgs/flagos-ai/repos?per_page=100&sort=updated |
| 12 | commit search (55 within window) | https://api.github.com/search/commits?q=org:flagos-ai+committer-date:%3E2026-08-20T02:18:00Z |
| 13 | Moore Threads H1 revenue RMB 1.736B, +147% (8/20, Huxiu) | https://news.google.com/rss/articles/CBMiVEFVX3lxTE9INkJCV2ViRnoyaVM5eFdrVWdlaTEzWmpCbTNUVEUwUS01QjA5S01sMGtBRGQyQW1YcGdQZHhiZ2tjQW1tXzd5ZzZHb1dQdVhWRHlBSA?oc=5 |
| 14 | Moore Threads Hong Kong IPO commentary (8/20, Caijinghao) | https://news.google.com/rss/articles/CBMia0FVX3lxTE1JTkdmYVNjQ0JiQlJMMllDbWo0RlNJQXhLaVVuU1YxNE1Da2VWVmxmT2RYTE5vSEljdGpnZXpLYXRuVzJXM3d2ZVdVZUNyUkhjMVpSY3ZneFUxNXdwMU9WWTNBckZkUUFoSzQ4?oc=5 |
| 15 | Hygon H1 revenue up over 60% (8/20, Sohu) | https://news.google.com/rss/articles/CBMiiAFBVV95cUxPWm5GWTVaNDlPS1NOX2lyM3ZQajVzbFh0R2VTaUhMZ05odTI4TllXcEtlQllQZGVZdnM3VE00NDRiVWV4RUppN21UT3BTNksyLVdNeUlSSW9xOElLR01URHNYano0REV6c244Y0p6cS0zTkJqcFNnSThub2RaTDJCaUNrbmY4bTh1?oc=5 |
| 16 | Hygon full-chain film and TV production (8/20, Sina Finance) | https://news.google.com/rss/articles/CBMif0FVX3lxTE54NllFbno4U1N2b3R6ZWw0b2F6d1UzaVhiTmRJXzhHUnF3ZXFqNTZBN0JEVWtfNUJEeUtxd0NZdDk0WVVDT2FZT3lWRGkyU1BwTUwxd1pTSk5oeG12M3RKZjhxTS1LN25aX1B5aDQ2bTZYRmtFNEdoRlkydnJ5VGM?oc=5 |
| 17 | Hygon Hefei token economy (8/20, Sina Finance) | https://news.google.com/rss/articles/CBMif0FVX3lxTE0zSXAxOUlMTXhUbGRfMnR0b2lac216c2lrVUI0NVRuaVBIOGE0bjdkUnY3bzlJWlFOQkd2MHE0Zk5lTWRIT2ROX3NQVllUbEdNeG1ESF9iUHcxaVFWNzFqb3FQaHNjejRsTmxuWnQ0dWpjMElGaFRRUE91RWwxZ2M?oc=5 |
| 18 | Hygon GIS ecosystem (8/20, Sohu) | https://news.google.com/rss/articles/CBMijAFBVV95cUxNUktpdVVZSG50VkZaM21yUk4zbWdWTFAwMEN2SmRZNWxIVnpfbG5QWmdzZk5tY1RLcWJFQXZUVjBrX0wwNWlGZkRMSGZmd3JzckdUZVI1U01jcnpQMDFmZUF5bWhsakRBZU5Bak5EcmN5Sm1JNFBZMEI4Qndta09oQ1lINEZFU2ZvcEJmVA?oc=5 |
| 19 | Cambricon H1 net profit RMB 2.311B, +122% (8/20, Guandian) | https://news.google.com/rss/articles/CBMiYkFVX3lxTFBXT2huRklrcS1qUmNYblpudGY3MlA4dDg2WWk0UFp3MDh4UWxIaXRmY1BuX0EzVzVKeXBxUFlnaHdyTWp3TWw4Q0VYMnJ4bV9uQ2hOUnNWUm42TTQtREtINGJn?oc=5 |
| 20 | Cambricon shares fall below RMB 1,000 (8/20, Guandian) | https://news.google.com/rss/articles/CBMiTkFVX3lxTE95eWJvX2tfZ01UUHNMRVpwX2ZCNmd5cnliLTZlVFEtZmtELWRuUWs0WlJYdURCSTJLaVgwOV9DRjJrU0JkVEJkUVhfWXV2Zw?oc=5 |
| 21 | Xingyun Technology wins MetaX “Industry Ark” general distribution rights (8/20, Sina Finance) | https://news.google.com/rss/articles/CBMilwFBVV95cUxQZ3poa0lobTBEbmVBUVliSzM2WHBJSTl4R1VaSVJKd2VpTDdsUmhVNUNTX0FmeGlMYVY5QkY4WmtfckEtc0k1bk1zaWVVbTBKQXFsdnN6YzkydmRZZVJVS0RfbzNXRDVRM1g2QzJ1R2h1U3pJV3l0Tnc5U2E4ZVpSbTdJWWRJQ3hQcWx0cGl3LTNQSlNFNXJN?oc=5 |
| 22 | Cloudpoint MetaX chain reaction (8/20, Sina Finance) | https://news.google.com/rss/articles/CBMieEFVX3lxTE1KNDlDS01seHJ1OWtlV0pLRVhwbGF6Z3hRZG9JcXM5dUlGUE53bkpINTdFMklPVEU4YWZLRUNqNGtUa05ZTHNEY0lJaEZlUy1uQTBzUkhkQ2VNTF9jRWdGblluY2t4T3BDQWRfWk4xZVhIYXg2N1NuYw?oc=5 |
| 23 | BAAI Community Meet AI Compiler #9 event recap (8/20) | https://hub.baai.ac.cn/view/57269 |
| 24 | Google News Chinese/English component query (zero hits) | https://news.google.com/rss/search?q=FlagOS+when:3d |
| 25 | HN Algolia | https://hn.algolia.com/api/v1/search_by_date?query=FlagOS |