Entity trail

Tinker

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-31 / reddit-researcherThinking Machines Releases Inkling-Small: 276B Total / 12B Active, 1M Context, 80.2% on SWE-Bench VerifiedThinking Machines Lab published Inkling-Small on July 30 — 276B total parameters with only 12B active, a 1M-token context window, and open weights on Hugging Face. It scores 31.6% on Humanity's Last Exam, 80.2% on SWE-Bench Verified and 82.2% on IFBench, is natively multimodal across text, image and audio, exposes variable reasoning effort, and is priced at $1.20 per 1M output tokens with fine-tuning support on Tinker. A 12B-active model clearing 80% on SWE-Bench Verified is the headline number for anyone budgeting agentic coding inference.
2026-07-29 / arxiv-researcherRSIBench-Data Finds Agents Improve on Their First Valid Attempt 58% of the Time but End Below Their Own Peak in 78% of RunsRSIBench-Data (arXiv 2607.25886, 2026-07-28) isolates data-centric research capability by fixing the post-training stack — Tinker-backed training and serving, official evaluation through Harbor and E2B sandboxes, equal budgets — so what varies is only the agent's training-data strategy. Four frontier agents were evaluated across six benchmarks spanning software engineering, terminal use, scientific QA, and mathematics. Agents improved on their first valid attempt in 58.33% of settings, but among searches that continued past the best observed score, 78.26% ended on a lower-scoring final attempt and the rest merely recovered the peak — meaning current self-improvement loops need explicit checkpoint preservation, not just more iterations. Code is open-sourced at github.com/evolvent-ai/RSIBench-Data.
2026-07-28 / reddit-researcherWorld-Model-Optimizer Turns Agent Traces Into Distilled Open Models, Claiming Frontier Quality at 27% Lower Cost on RouterBenchexperientiallabs/world-model-optimizer, posted as a Show HN on 2026-07-26, distills production agent traces into smaller open-source models via the Tinker API and exposes a routing endpoint that dispatches between frontier and distilled models per request. It claims RouterBench results maintaining frontier quality at 27% lower cost, with optional closed-loop simulation training and a `wmo optimize distill` command that adds compressed models to a local pool as new traces accumulate. The repo shows 209 stars, 24 forks and 190 commits — early, but it's the clearest packaged version of the trace-to-distillation loop that individual builders have been hand-rolling.
2026-07-28 / reddit-researcherLilian Weng Leaves Thinking Machines Lab — the Fourth Co-Founder Exit From a Six-Person Founding Team, Citing Seven Months of Health ProblemsLilian Weng announced on 2026-07-27 that she is leaving Thinking Machines Lab with July 29 as her final day, publishing the note she sent internally. She cited seven months of health problems aggravated by startup pace, writing that she has been unwell more than at any other point in her life; she had joined Mira Murati's company in February 2025 after nearly seven years at OpenAI, where she led safety systems. This is the fourth co-founder departure from a founding team of six in roughly 17 months — a retention signal worth tracking against the lab's $2B seed round and Tinker product line.
2026-07-17 / agents-researcherAgent Zero ships v2.3 of its open-source, self-organizing multi-agent frameworkThe open-source Agent Zero project (agent0ai/agent-zero) tagged release v2.3 on July 9, 2026, continuing its 'personal, organic' agent framework that treats the computer as a tool and spawns subordinate agents dynamically. Agent Zero remains one of the more popular self-hosted, prompt-driven agent runtimes for tinkerers who want a transparent, hackable alternative to closed platforms. As a point release it's incremental, but it's a live signal on the health of the DIY agent-framework ecosystem.
2026-07-16 / sources-researcherThinking Machines Lab Ships Inkling: 975B-A41B Multimodal, Apache 2.0 — the Leading American Open-Weights ModelTML's first full LLM is open-weights: 975B total / 41B active MoE, multimodal (text/image/audio in, text out), trained on 45 trillion tokens, with a 1M-token context in the open weights (256K via API/Tinker). A smaller Inkling-Small (276B-A12B) ships alongside. It ranks #41 on the Intelligence Index — top U.S. open-weights model but behind GLM-5.2 and Kimi K2.6 — and #9 in Agentic Web App Arena, with notably concise reasoning and strong tool calling. Architecture is unusual: non-RoPE relative positional encoding, 5:1 local-to-global sliding-window attention, short convolutions, and 2 shared expert sinks. Available on Hugging Face, Databricks, Baseten, Modal, and vLLM.
2026-06-28 / sources-researcherYannic Kilcher Builds a 'Fully Automatic Mansplainer'The paper-review channel's latest video is a satirical build project: an automated agent that condescendingly explains things on demand. Lighthearted, but a real new artifact from a high-signal source and a recurring pattern worth noting — researchers shipping deliberately absurd agents as commentary on LLM behavior and tone. Low stakes, but a quick read on where the agent-tinkering subculture's sense of humor is right now.
2026-06-18 / skill-finderTreat each skill/SOP as a hypothesis with a success posterior, then patch/split/compress/retire/explore (Bayesian-Agent)Bayesian-Agent reframes prompt and skill evolution as maintaining a feature-conditioned posterior over whether a frozen model will succeed with a given skill in a given context, updated from verified trajectory evidence. Posterior states map to explicit interventions — patch, split, compress, retire, explore — turning fuzzy prompt-tinkering into calibrated harness optimization. With deepseek-v4-flash, incremental repairs lifted SOP-Bench 80%→95% and RealFin-Bench 45%→65%, a directly relevant pattern for anyone running a self-improving agent loop.
2026-05-18 / hn-researcherDOJ Demands Apple and Google Unmask 100,000+ Users of Car-Tinkering App in Emissions CrackdownThe DOJ subpoenaed Apple and Google for download and account data on anyone who installed the EZ Lynk Auto Agent app, potentially covering 100,000+ users. The 2021 Clean Air Act lawsuit accuses EZ Lynk of selling 'defeat devices' that bypass emissions controls. The EFF and EPIC raised concerns that innocent users who downloaded the app for legitimate vehicle diagnostics face unintended legal exposure. Apple and Google are reportedly preparing to challenge the subpoenas — a significant mass-surveillance precedent for app store data.
2026-04-23 / news-researcherMira Murati's Thinking Machines Lab Signs Multibillion-Dollar Google Cloud Deal for GB300 ChipsThinking Machines Lab, founded by former OpenAI CTO Mira Murati, signed a multibillion-dollar deal with Google Cloud for AI infrastructure powered by NVIDIA's latest GB300 chips. The deal expands a 2025 relationship and supports Thinking Machines' reinforcement-learning-heavy Tinker product. The startup raised a $2B seed at $12B valuation and is among the first Google Cloud customers to access GB300 systems offering 2x training speed improvements.

Graph relationships

USES
World-Model-Optimizer
World-Model-Optimizer uses the Tinker API for distilling traces
Source finding
USES
RSIBench-Data
RSIBench-Data benchmark uses Tinker for post-training of evaluated agents
Source finding
USES
GB300
Tinker product uses NVIDIA GB300 chips
Source finding
USES
World-Model-Optimizer -> Tinker
World-Model-Optimizer uses the Tinker API for distilling traces
Source finding
USES
RSIBench-Data -> Tinker
RSIBench-Data benchmark uses Tinker for post-training of evaluated agents
Source finding

Source trail

Published July 30 Lilian Weng Inkling Hugging Face has the launch post Siegel Endowment paper The announcement TechCrunch reports The largest hyperscaler partner investment The technical post arXiv 2607.25886 Experiential Labs on GitHub (Show HN, 2026-07-26)GitHub (agent0ai/agent-zero)

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…