TinyFish
Sign up
Products
MakoWeb-native model for live workflows
TinySearchFast, structured web search
TinyFetchAny URL to clean content
TinyAgentMulti-step web automation
TinyBrowserCloud browser sessions
Resources
DocumentationAPI reference and guides
PartnersThe Data Partner Alliance
IntegrationsConnect with your stack
CompareTinyFish vs alternatives
BenchmarksHow we measure up
BlogProduct updates and insights
PulseWhat we've shipped
CookbookOpen-source examples
PricingRates, Wallet, and billing
Enterprise
OverviewWeb execution for enterprise teams
SolutionsWhat we run, and where it runs in production
CustomersSee who builds with TinyFish
ContactLog In
Together

What actually changed in AI this year?

Try it for free
TinySet prompt
Every frontier AI model released this year, with lab, release date and what's new about it
tinyfish-what-actually-changed-in-ai-this-year.csv90 rows
Model nameLabRelease dateWhat's newAnnouncement URLIs frontier
Gemini 4 CarbonGoogle DeepMind2026-10-28Google DeepMind's strongest model, with state-of-the-art cybersecurity capabilities and long-horizon agentic/coding performance.https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/Yes
GPT-6.1 TerraOpenAI2026-10-26Fast, efficient model combining detailed visual reasoning and precise computer use with clearer writing; 50% cheaper API prices than GPT-5.6 Terra, over 3x faster than GPT-6 Astrahttps://openai.com/index/introducing-gpt-6-1-terra/Yes
Kimi K3.1 ProMoonshot AI2026-10-23First model in Moonshot's K3.1 series and its most capable 3T-class open-weight model, with a 1M-token context window and substantial gains in coding, 3D creation, financial research, and agentic capabilities.https://www.kimi.ai/blog/kimi-k3-1-proYes
MAI-Base-2Microsoft AI2026-10-15An MoE foundation model with 80 billion active and 2.8 trillion total parameters that outperforms every open-weight model tested on held-out evaluations.https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/Yes
Nemotron 3.5 Super VLNVIDIA2026-10-15An open 120-billion-parameter multimodal mixture-of-experts model delivering the fastest task completion and highest accuracy for specialized always-on agents, combining visual reasoning, tool use and code generation for agentic systems.https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/Yes
Muse Spark OpenMeta2026-10-15Meta's most capable US open-weight model to date: a natively multimodal 816B-parameter (42B active) mixture-of-experts with a 1M-token context window, built for coding, agentic work and long-horizon reasoning.https://research.meta.ai/blog/muse-spark-openYes
Gemini 4 ArgonGoogle DeepMind2026-10-13Google's latest frontier model, it excels at deep work in complex, real-world tasks across coding, finance, legal, cybersecurity, and science, ranking #1 on the Vals Index.https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/Yes
Muse WonderMeta2026-10-12Meta's most capable model to date, delivering state-of-the-art performance on personal and professional agentic tasks with Active Memory (carrying context across conversations) and Active Focus (persisting through hard tasks), available via the Meta Model API and Muse Code.https://about.fb.com/news/2026/10/introducing-muse-wonder-models/Yes
Showing 8 of 90 rows
Download dataset (CSV)Try this prompt in TinySet

Then TinyResearch on the dataset:

TinyResearch prompt
What actually changed in AI this year? Focus on what mattered and skip the hype.
tinyfish-what-actually-changed-in-ai-this-year.pdf

2026 was the year AI moved from impressive demos to genuine capability jumps — but also the year the gap between benchmark headlines and real-world deployment became impossible to ignore. The most consequential developments: frontier models crossed into research-grade scientific work for the first time; AI coding agents saturated their original benchmarks and required harder ones; Chinese open-weight models narrowed the gap with U.S. frontier labs to a few months while dramatically undercutting on price; global AI infrastructure investment approached $1 trillion; and regulatory frameworks either slipped deadlines (EU) or remained largely voluntary (U.S.). Enterprise adoption spread widely but scaled poorly, with most pilot programs failing to reach production. Safety practices improved at the margins while companies simultaneously weakened earlier self-imposed limits. The net judgment: AI's capability frontier moved decisively forward; the institutional infrastructure to govern and deploy it did not keep pace.

Download full report (PDF)Try this prompt in TinyResearch

Related

Together

What is YC betting on in AI right now?

153 rows · YC has made its most aggressive, explicit pivot in its two-decade history. AI is no longer a theme within batches — it is the substrate. Roughly 95% of the Spring 2026 batch touched AI, and across 2026 batches approximately 60% of companies are explicitly AI-focused, up from ~40% in 2024. The directional shift visible in the Fall 2026 Request for Startups (RFS) — 13 categories, framed as "AI moving into the physical world" — is the clearest signal of where YC is placing chips next: physical infrastructure, defense, human-AI collaboration frameworks, and the "ownership" of intelligence rather than renting it. The December 2 Demo Day will reveal the F26 cohort fully, but early company data and the RFS together paint a sharp strategic picture right now.

Run this prompt
Together

Which LLM API gives the best value for money?

105 rows · The LLM API market in October 2026 is sharply bifurcated: a cheap tier ($0.10–$0.75/M input tokens) that now delivers capability matching premium models of 18 months ago, and a frontier tier ($4–$10/M input) for the highest-stakes agentic and reasoning work. The single most important finding is that the efficiency frontier has moved decisively: GPT-6.1 Sol at $2/$10 scores 51.8 on the Artificial Analysis Intelligence Index and Claude Sonnet 5.5 at $2/$10 scores 47–56 depending on effort level, while Claude Haiku 5.5 and GPT-6 Luna deliver scores of roughly 28–43 at just $0.10/$0.50—a price point that was frontier-model territory less than two years ago.

Run this prompt
Together

Which AI conferences in 2027 are worth the ticket?

125 rows · The 2027 AI conference landscape divides cleanly into four categories: (1) flagship academic research conferences (NeurIPS, ICML, ICLR, CVPR, AAAI, ACL, ICCV) that are peer-reviewed, selective, and best for researchers and technically senior engineers tracking the frontier; (2) applied/builder conferences (AI Engineer World's Fair, MLSys, Databricks Data+AI Summit, NVIDIA GTC) that are best for engineers shipping production AI systems; (3) commercial/executive summits (RAISE, AI Summit London, HumanX, Ai4) best for C-suite, founders, enterprise buyers, and investors; and (4) subfield-specialist venues (IROS, RSS for robotics; ACL/NAACL/EACL for NLP; CVPR/ICCV/WACV for vision; FAccT for AI ethics) that serve narrow technical communities.

Run this prompt