Sureform builds post-training datasets and reinforcement-learning environments grounded in real enterprise workflows to improve frontier agents. The Y Combinator–backed team focuses on multimodal human data collection starting with video for AGI-oriented model training.
SWE-bench is a family of open software-engineering benchmarks built from real GitHub issues, with public leaderboards comparing model and agent performance. The site tracks variants such as Verified, Lite, Multilingual, and Multimodal tasks and links related tools like mini-SWE-agent and SWE-agent.
Synapse Semiconductor is a frontier semiconductor lab building sensing and compute hardware for physical AI. The team focuses on chips that understand and filter visual input at the sensor rather than shipping full frames to remote GPU compute.
TabPFN from Prior Labs is a pre-trained tabular foundation model for fast classification, regression, forecasting, and related structured-data tasks without dataset-specific training loops. Teams can try TabPFN via API, private cloud, or on-prem deployments with open-source and commercial licensing options.
TabTabTab builds machine intelligence models focused on closing the gap between software that thinks and software that acts, validating systems on real hardware. The public site is a sparse landing page with positioning copy rather than a commercial SaaS catalog.
Tafi, offered through Daz 3D, supplies commercially licensed structured 3D character datasets for training AI models. The program emphasizes clean topology, rigged assets, automatic labels, and scalable synthetic data generation for gaming, computer vision, and virtual human use cases.
Tavus builds real-time conversational video experiences through developer APIs and no-code PAL tools. It offers stock and custom AI replicas, multilingual voice and video calls, memory and knowledge bases, and models such as Phoenix, Raven, Sparrow, and Griffin for human-like interactions.
TensorFlow Playground is a browser-based neural network sandbox for tinkering with datasets, activations, regularization, and training dynamics in real time. It helps learners visualize how hidden layers, weights, and hyperparameters affect classification and regression problems.
The Agentic Data Company designs and collects real conversational audio datasets for frontier speech-model training. It records consented, licensed dialogue at lab scale—including accents, turn-taking, and multi-party settings—and publishes datasets such as Open Yap 1K while building custom collections for model teams.
The Forecasting Company builds foundation models for time series, starting with t0-beta for universal forecasting. Its API accepts raw historical data without per-domain fine-tuning, returns calibrated future distributions, and supports large-scale backtests plus open weights and interactive Retrocast demos.
Theora (Louiza, Inc.) builds analyst-grade AI for biopharma by teaching models how real clinical work unfolds. It creates healthcare datasets, expert reasoning traces, and FHIR-native EHR and PACS simulation environments so agents can master multi-step medical tasks with verifiable rewards.
The Pile is an 825 GiB open-source language modeling corpus from EleutherAI combining 22 high-quality sub-datasets for training large models. It ships as zstandard-compressed jsonlines hosted by the Eye with an accompanying arXiv paper.
Theta builds specialized AI agents by translating real-world workspaces into simulation environments. It synthesizes first-party and expert data into custom environments, trains agent workforces on long-horizon tasks, and delivers forward-deployed solutions for frontier labs and enterprises.
The Token Company sells bear-2 prompt compression that strips filler tokens before requests reach frontier LLMs. Its deterministic models cut input size in tens of milliseconds, preserve answer quality in customer case studies, and offer Pro and Enterprise deployments with trust-center controls for regulated data.
Tinfoil builds verifiably private AI using secure hardware enclaves. It offers Private Chat, an OpenAI-compatible Private Inference API, and Tinfoil Containers for running Docker workloads with attested privacy, plus SOC 2 compliance and open-source components.
Together AI is the AI-native cloud for open-source model inference, fine-tuning, GPU clusters, and research tooling. It offers serverless and dedicated inference, batch jobs, custom training, evaluations, and a large library of chat, code, image, and video models.
Tokenless finds flaws in reinforcement-learning environments, evals, and graders that models exploit as reward hacking. The Y Combinator-backed team offers tooling and expertise to harden tasks so training signal reflects real capability gains.
TokenMix is an OpenAI-compatible AI API relay operated by Token Limited (Hong Kong) that routes chat, image, video, audio, and embedding models through one base URL and prepaid wallet. Developers top up from $1, pick model IDs from a live catalog, and monitor usage, keys, and provider status in a dashboard without juggling separate vendor accounts.
Touchmark runs a live order book for reserved AI inference capacity, letting buyers commit to future token delivery months and providers sell forward GPU output. Contracts pin model-specific token blocks with curve-based discounts up to twenty percent for early commitments and OpenAI-compatible drawdown keys.
Traceloop provides observability, evaluation, and prompt management for LLM applications with OpenLLMetry instrumentation. Pricing advertises a free forever tier for up to fifty thousand spans monthly plus enterprise deployment options, and the company notes it is joining ServiceNow.
Tracer is a Y Combinator-backed research lab building coordinated ensembles of specialist open models. Its Echo product offers one OpenAI-compatible endpoint that adaptively routes work across models to improve performance per token.
Not listed
API available
Checked Oct 4, 2026
Showing 600 of 664 tools
Plan facts carry individual sources and check dates. Unknown values are excluded from required filters.