Neolabfresh 18h

IBM (Granite)

www.ibm.com/granite ↗github/ibm-granite ↗dossier.jsonrss
Signal timeline282 total
Oct 28, 2026
1dModelibm-granite/LLMON-BenchmarksBenchmark dataset release, no major traction info.sourcenotability 5.0/10
Oct 14, 2026
2wModelibm-granite/granite-switch-4.1-8bIBM model release but very low downloadssourcenotability 5.0/10↓3.6k♥54
2wModelibm-granite/granite-switch-4.1-30bNotable IBM model releasesourcenotability 7.0/10↓2.6k♥40
Oct 7, 2026
3wModelibm-granite/granite-speech-5.0-2bIBM Granite speech model release, low tractionsourcenotability 5.0/10↓4k♥65
Oct 1, 2026
3wModelibm-granite/ensemble-forecasting-with-ibm-granite-time-seriesSolid ensemble forecasting repo by IBMsourcenotability 5.0/10♥1
Sep 23, 2026
Sep 23Modelibm-granite/granite-timeseries-ensemble-r1Low traction model release with only 25 downloads.sourcenotability 4.0/10↓320♥8
Sep 1, 2026
Sep 1Modelibm-granite/granite-4.2-30b-bf16-mlxIBM released Granite 4.2 30B MLX variant, notable enterprise model.sourcenotability 7.0/10↓563♥1
Sep 1Modelibm-granite/granite-4.2-8b-bf16-mlxFormat variant of existing notable model, solid reposourcenotability 5.0/10↓574♥1
Sep 1Modelibm-granite/granite-4.2-3b-bf16-mlxOfficial MLX variant of IBM Granite 3B modelsourcenotability 6.0/10↓512♥1
Aug 17, 2026
Aug 17Modelibm-granite/granite-speech-5.0-470m-turboctc-ncLow traction model release from IBM, small size.sourcenotability 5.0/10↓38k♥46
Aug 13, 2026
Aug 13Modelibm-granite/granite-4.2-8b-nvfp4Notable IBM Granite 4.2 8B quantized model releasesourcenotability 7.0/10↓21♥3
Aug 13Modelibm-granite/granite-4.2-8b-mxfp4Notable quantized 8B model release from IBMsourcenotability 6.0/10↓23
Aug 13Modelibm-granite/granite-4.2-8b-fp8Quantized IBM Granite model release.sourcenotability 6.0/10↓348
Aug 13Modelibm-granite/granite-4.2-3b-nvfp4routine quantized model release, negligible downloadssourcenotability 3.0/10↓15♥1
Aug 13Modelibm-granite/granite-4.2-3b-mxfp4IBM released Granite 4.2 3B quantized variantsourcenotability 6.0/10↓13♥1
Aug 13Modelibm-granite/granite-4.2-3b-fp8Official FP8 quantized variant of existing small model.sourcenotability 5.0/10↓71♥2
Aug 13Modelibm-granite/granite-4.2-30b-nvfp4IBM Granite 4.2 30B NVFP4 model release, notable enterprise LLMsourcenotability 7.0/10↓50♥2
Aug 13Modelibm-granite/granite-4.2-30b-mxfp4Low traction quantized model releasesourcenotability 3.0/10↓11♥2
Aug 13Modelibm-granite/granite-4.2-30b-fp8IBM quantized model variant, minimal downloads.sourcenotability 4.0/10↓18♥2
Aug 7, 2026
Aug 7Modelibm-granite/granite-4.2-30bIBM 30B model release, low community tractionsourcenotability 6.0/10↓36k♥137
Aug 7Modelibm-granite/granite-4.2-8bIBM releases Granite 4.2 8B, modest traction.sourcenotability 5.0/10↓73k♥110
Aug 7Modelibm-granite/granite-4.2-3bIBM Granite 4.2 3B release, low tractionsourcenotability 5.0/10↓40k♥118
Aug 7Modelibm-granite/granite-timeseries-patchtst-fm-r2IBM Granite time series foundation model releasesourcenotability 6.0/10↓212k♥17
Aug 4, 2026
Aug 4Modelibm-granite/granite-speech-5.0-470m-turboctcLow-download IBM model release, not major.sourcenotability 5.0/10↓62k♥111
Jul 1, 2026
Jul 1Modelibm-granite/granite-swash-3b-a600mIBM's open model release with modest downloads, not flagship.sourcenotability 5.0/10↓7k♥16

Top signals

  1. #1WritingIBM and Qedma Demonstrate Quantum Advantage, Modeling Physics Beyond Classical Capabilities Through Trusted Quantum Computation8.0
  2. #2WritingIBM Commits More Than $10 Billion to Quantum Computing, Funding Its Roadmap from Today's Leading Systems to the World's First Fault-Tolerant Quantum Computers8.0
  3. #3WritingIBM Completes Acquisition of HRL Laboratories to Accelerate the Future of Quantum8.0
  4. #4WritingIBM Advances Enterprise AI Software Development with Multi-Agent Capabilities and Specialized Modernization Workflows7.0
  5. #5WritingIBM and NASA Release Open-Source AI Model to Support Lunar Exploration7.0

Agent answer

IBM (Granite) has 282 loaded public signals: 0 hiring, 0 forks, 190 releases or model cards, 63 talking, and 29 repos. Latest signal: USTA and IBM Announce Five-Year Renewal of Longstanding Partnership . Data-business radar is currently scoped to frontier labs, so this category does not expose radar lanes. The standing analysis was generated with deepseek-v4-pro and 94 evidence refs.

Hiring 0Forks 0Releases 190Talking 63Repos 29
IBM (Granite)

has loaded 282 public signals

IBM (Granite)

has hiring signal count 0

IBM (Granite)

has fork signal count 0

IBM (Granite)

has release signal count 190

Analysis — agent synthesisfull report →generated October 29, 2026

Thesis

IBM's Granite program reads as an enterprise-open model operation converging on three bets: (1) small, modular, Apache-2.0 models that pack chat, decision, safety-guardian, and speech into single endpoints P13W1; (2) an "AI operating model" for governing agentic AI — identity, least-privilege access, and evaluation — rather than simply shipping more models P7P15; and (3) open inference and security infrastructure (llm-d, Lightwell) aimed at lowering the cost and risk of production AI P12P21. The earnings narrative frames this as "Super Intelligence" transforming operating models, with Software and Infrastructure leading revenue growth P11. Unlike model-only labs, IBM's signals span models, semiconductors (nanostack), and quantum, indicating a diversified compute roadmap beyond LLMs P16P17.

Signal desks

Hiring

  • IBM Research is recruiting researchers and engineers for its Core Foundation Models team in the United States and India, with listed interests in foundation models, reinforcement learning, agentic AI, and enterprise AI systems W5.
  • Adjacent scaling via acquisition: IBM acquired Logiq Consulting to expand UK cybersecurity and sovereign-technology capabilities for regulated and public-sector clients E30. No other role-level hiring evidence is present in this pack; the hiring desk is thin beyond W5.

Forks

  • No cited evidence in this pack.

Releases

  • Granite 4.2 line (3B/8B/30B, Apache-2.0, text-generation) E1E2E3, plus BF16 MLX variants for 3B/8B/30B E39E40E41.
  • Granite Speech 5.0 (470m-turboctc, 2b) E4E5P9, including a non-commercial cc-by-nc-sa turboctc variant E8.
  • Granite Switch 4.1 (8B/30B) composing a base model with Decision and Guardian adapters plus a speech encoder E6E7P13P14.
  • Time-series: Granite Timeseries Ensemble R1 and patchtst-fm-r2 E19E27; granite-tsfm tooling releases v0.3.9 → v0.3.11 E44E25E15.
  • LLMON-Benchmarks model card/repo E10P6; gguf test tags (test-vX-language-01→08; test-v5.0-oss-120B-04→13) signal a 120B "Granite 5.0 OSS" checkpoint in pre-release testing E46E55.

Talking

  • Governance/security: agent identity and control via HashiCorp Vault and watsonx Orchestrate P7E11; EMEA study on the agent-governance gap P15E17; Lightwell 400+ vulnerability remediation P21E12.
  • Infrastructure/sovereignty: llm-d 1.0 (CNCF-governed) P12E16; self-hosted IBM Bob P26E28; Swift integration and Digital Asset Haven on-premises P28E31.
  • Commercial/GTM: USTA renewal through 2031 P5E9; Scuderia Ferrari fan app P8E13; SAP and Providence wins P20P18; Q3 results P11E14.
  • Research breadth: DARPA Quantum Benchmarking Stage C P17E20; Siemens nanostack collaboration P16E18; NASA lunar model P2E35; HRL Laboratories acquisition E50; Gordon Bell quantum protein simulation E38.

Shipping

  • Granite 4.2 (3B/8B/30B) shipped August 2026 under Apache-2.0 E1E2E3; the 4.2 reasoning models were pre-trained from scratch on roughly 15T tokens with a five-phase strategy extending the context window and sandboxed tool-use W3.
  • Speech: granite-speech-5.0-2b (multilingual ASR/AST for English, French, German, Spanish, Portuguese, Japanese; trained on 390,000 hours) P9, plus the 470m-turboctc model E4 and a non-commercial nc variant E8; the launch blog confirms three simultaneous updates W2.
  • Switch 4.1: one endpoint for chat, decisions, guardian safety, and speech at roughly base-model VRAM W1P13.
  • Time-series: ensemble framework plus patchtst-fm-r2 (212,338 downloads) E19E27P22; granite-tsfm 0.3.10/0.3.11 added ensemble support and packaging P23P10.
  • Eval/benchmark artifact: LLMON-Benchmarks E10P6.
  • Pre-release signal: gguf "test-v5.0-oss-120B" tags imply a 120B Granite 5.0 OSS checkpoint is in flight E55E56.

Research themes

  • Modularity: aLoRA adapters (Decision, Guardian) composed onto a base model via control tokens P13W1; Decision adapters added to granitelib-core W4.
  • Speech/audio: accuracy, noise/reverb robustness, keyword biasing, and language-ID prompting P9W2.
  • Reasoning/agents: 15T-token pretraining, five-phase training, and sandboxed tool-use W3; agentic AI is also an explicit hiring interest W5.
  • Time-series/forecasting: ensemble of PatchTST/FlowState/TTM checkpoints evaluated with GiftEval P22.
  • Scientific AI and compute: NASA-IBM Lunar Foundation Model P2E35; 12,635-atom quantum protein simulation (Gordon Bell finalist) E38; fault-tolerant quantum roadmap (2029 target, Qiskit, 340+ orgs) P17; nanostack 3D transistors at 7 Å P16.

Hiring & scaling

  • Direct role evidence is limited to IBM Research's Core Foundation Models team recruiting in the US and India for foundation models, RL, agentic AI, and enterprise AI W5.
  • Scaling is partly via M&A and facilities: HRL Laboratories (quantum/materials) E50, Logiq Consulting (UK cybersecurity/sovereign) E30, and Anderon's $1B CHIPS award to scale a U.S.-based pure-play quantum foundry E34.
  • Commercialization buildout: self-hosted IBM Bob targets on-prem, sovereign-cloud, and air-gapped software development P26; watsonx Orchestrate unifies agent discovery, containment, and evaluation across Amazon Bedrock, Microsoft Foundry, and Google Gemini P7.

Category implications

  • Strategy: IBM is selling an "AI operating model" — governed identity, least-privilege access, and evaluation for agents — using the governance gap (80% incident escalation vs 40% policy change) as the wedge P7P15.
  • Infrastructure: llm-d 1.0 (CNCF-governed; founded with Red Hat, Google Cloud, CoreWeave, NVIDIA) targets cost-efficient distributed inference, with claims of 5–10x cost reductions on context-heavy agentic workloads P12; Lightwell converts open-source vulnerability discovery into backported fixes P21.
  • Product: modular Switch = capability density in one endpoint W1; self-hosted Bob addresses sovereignty and governance constraints P26; Digital Asset Haven/Swift extends regulated-industry GTM P28.
  • Research: quantum (DARPA QBI Stage C, HRL acquisition, CHIPS award) and semiconductor (nanostack + Siemens EDA) show a compute-research roadmap beyond LLMs P17E50E34P16.
  • Hiring: foundation-model + RL + agentic hiring in the US and India signals continued in-house model capability W5.
  • GTM: sports partnerships (USTA through 2031; Ferrari) serve as public demo surfaces for hybrid cloud/AI P5P8; the EMEA-governance, CFO, and K-12 studies function as market-education GTM P15P27P3.

Traction highlights

  • Hugging Face downloads: granite-4.2-8b 72,936 E3; granite-4.2-3b 39,863 E2; granite-4.2-30b 36,188 E1; granite-speech-5.0-470m-turboctc 61,527 E4; granite-timeseries-patchtst-fm-r2 212,338 E27.
  • Business: Q3 revenue $17.1B (+5%), Software +8%, full-year free-cash-flow outlook raised by more than $1B P11.
  • Platform: 270M+ total engagements across USOpen.org and the US Open app; partnership renewed through 2031 P5.
  • Adoption claims: llm-d "broadly adopted by frontier model labs, hyperscalers, neoclouds" P12; quantum computers used by a network of 340+ organizations P17.
  • Discussion traction is thin: the only HN metric captured in this pack is 2 points/0 comments on the Lightwell post E12; engagement data is absent for most other posts.