Marktechpost has launched AI2025Dev, its 2025 analytics platform (accessible to AI Devs and Researchers with none signup or login) designed to transform the yr’s AI exercise right into a queryable dataset spanning mannequin releases, openness, coaching scale, benchmark efficiency, and ecosystem contributors. Marktechpost is a California primarily based AI information platform masking machine studying, deep studying, and information science analysis.
What’s new on this launch
The 2025 launch of AI2025Dev expands protection throughout two layers:
- Launch analytics, specializing in mannequin and framework launches, license posture, vendor exercise, and have degree segmentation.
- Ecosystem indexes, together with curated “Prime 100” collections that join fashions to papers and the folks and capital behind them. This launch contains devoted sections for:
- Prime 100 analysis papers
- Prime 100 AI researchers
- Prime AI startups
- Prime AI founders
- Prime AI traders
- Funding views that hyperlink traders and corporations
These indexes are designed to be navigable and filterable, somewhat than static editorial lists, so groups can hint relationships throughout artifacts like firm, mannequin kind, benchmark scores, and launch timing.
AI Releases in 2025: yr degree metrics from the market map dataset
AI2025Dev’s ‘AI Releases in 2025’ overview is backed by a structured market map dataset masking 100 tracked releases and 39 lively firms. The dataset normalizes every entry right into a constant schema: title, firm, kind, license, flagship, and release_date.
Key mixture indicators on this launch embrace:
- Whole releases: 100
- Open share: 69%, computed because the mixed share of Open Supply and Open Weights releases (44 and 25 entries respectively), with 31 Proprietary releases
- Flagship fashions: 63, enabling separation of frontier tier launches from spinoff or slim scope releases
- Lively firms: 39, reflecting a focus of main releases amongst a comparatively fastened set of distributors
Mannequin class protection available in the market map is explicitly typed, enabling faceted queries and comparative evaluation. The distribution contains LLM (58), Agentic Mannequin (11), Imaginative and prescient Mannequin (8), Device (7), Multimodal (6), Framework (4), Code Mannequin (2), Audio Mannequin (2), plus Embedding Mannequin (1) and Agent (1).
Key Findings 2025: class degree shifts captured as measurable indicators
The discharge packages a ‘Key Findings 2025’ layer that surfaces yr degree shifts as measurable slices of the dataset somewhat than commentary. The platform highlights three recurring technical themes:
- Open weights adoption, capturing the rising share of releases with weights accessible beneath open supply or open weights phrases, and the downstream implication that extra groups can benchmark, nice tune, and deploy with out vendor locked inference.
- Agentic and gear utilizing programs, monitoring the expansion of fashions and programs categorized round instrument use, orchestration, and job execution, somewhat than pure chat interplay.
- Effectivity and compression, reflecting a 2025 sample the place distillation and different mannequin optimization strategies more and more goal smaller footprints whereas sustaining aggressive benchmark habits.
LLM Coaching Information Scale in 2025: token scale with timeline alignment
A devoted visualization tracks LLM coaching information scale in 2025, spanning 1.4T to 36T tokens and aligning token budgets to a launch timeline. By encoding token scale and date in a single view, the platform makes it attainable to check how distributors are allocating coaching budgets over time and the way excessive scale pertains to noticed benchmark outcomes.
Efficiency Benchmarks: benchmark normalized scoring and inspection
The Analytics part features a Efficiency Benchmarks view and an Intelligence Index derived from customary analysis axes, together with MMLU, HumanEval, and GSM8K. The target is to not exchange job particular evaluations, however to offer a constant baseline for evaluating vendor releases when public reporting differs in format and completeness.
The platform exposes:
- Ranked efficiency summaries for fast scanning
- Per benchmark columns to detect tradeoffs (for instance, coding optimized fashions that diverge from reasoning centric efficiency)
- Export controls to help downstream evaluation workflows
Mannequin Leaderboard and Mannequin Comparability: operational analysis workflows
To cut back the friction of mannequin choice, AI2025Dev contains:
- A Mannequin Leaderboard that aggregates scores and metadata for a broader 2025 mannequin set
- A Mannequin Comparability view that allows facet by facet analysis throughout benchmarks and attributes, with search and filtering to construct shortlists by vendor, kind, and openness
These workflows are designed for engineering groups that want a structured comparability floor earlier than committing to integration, inference spend, or nice tuning pipelines.
Prime 100 indexes: papers, researchers, startups, and traders
Past mannequin monitoring, the launch extends to ecosystem mapping. The platform provides navigable “Prime 100” modules for:
- Analysis papers, offering an entry level into the core technical work shaping 2025 programs
- AI researchers, offered as an unranked, proof backed index with convention anchored context
- AI startups and founders, enabling linkage between product course and launched programs
- AI traders and funding, enabling evaluation of capital flows round mannequin and gear classes
Availability
The up to date platform is obtainable now at AI2025Dev and also you don’t want any signup or login to entry the platform. The discharge is designed to help each quick scanning and analyst grade workflows, with normalized schemas, typed classes, and exportable views meant for quantitative comparability somewhat than narrative searching.
Asif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is dedicated to harnessing the potential of Synthetic Intelligence for social good. His most up-to-date endeavor is the launch of an Synthetic Intelligence Media Platform, Marktechpost, which stands out for its in-depth protection of machine studying and deep studying information that’s each technically sound and simply comprehensible by a large viewers. The platform boasts of over 2 million month-to-month views, illustrating its reputation amongst audiences.
