Updates
Release Notes
All updates and innovations on the AIGENCY platform.
4.1May 28, 2026
AIGENCY V4.1 — Folders, Deep Research, Knowledge Base and More
NewChat Folders are here. Create folders from the colored pill bar in the sidebar to organize your chats. When you delete a folder, chats inside move to "All" — nothing is lost.
NewDeep Research is live on Pro and Business plans. Enable Deep Research in the chat header, ask anything, and the assistant spends 5–10 minutes producing a multi-source report. It runs in the background; the result appears in your chat when ready.
NewKnowledge Base is here. Upload your own PDF, Word or text files and link them to an assistant — it answers your questions by citing your own documents. Open it from the "KB" button in the sidebar.
NewScheduled Tasks are here. Set up recurring tasks like "every morning at 9 summarize today's news" — the assistant runs automatically and drops the result in a new chat. Find it under "Tasks" in the sidebar.
NewCanvas — a side editor — is here. Click Canvas under any long reply; the editor opens on the right and you can select text and edit with AI via presets like "Make shorter", "More formal" or "Debug".
ImprovementRegenerate: not happy with an answer? Use the refresh button under the message to get a new answer to the same question. Daily limit varies by plan.
Read MoreWe wrote a full guide with screenshots, real-world scenarios and step-by-step usage on our blog: How to use 5 new AIGENCY V4.1 features
4.0April 27, 2026
AIGENCY V4 — Multimodal Model Launch
LaunchAIGENCY V4 is now in production. 120B sovereign text core + 8B sovereign vision encoder = 128 billion parameters total, fully independent and multimodal architecture. Technical whitepaper published with 13,344 real API calls, 22 benchmarks, Wilson 95% confidence interval.
MultimodalVisual input understanding added: document Q&A (DocVQA 79.17), chart interpretation (ChartQA 67.68), multimodal reasoning on university subjects (MMMU 53.33). Sovereign vision encoder fine-tuned on 8 million Turkish-captioned images.
Turkish LeaderWorld leader status reinforced in Turkish reading and natural-language inference: Belebele-TR 87.33, TQuAD 82.40, TR-MMLU 70.80, XNLI-TR 73.40, TR Grammar 79.00. No frontier model publishes Turkish-specific comparisons — V4 is the de facto global reference.
Frontier LevelSame band as frontier models on scientific reasoning and grade-school math: ARC-Challenge 94.88, GSM8K 94.62. Upper-mid segment on code generation: HumanEval 84.15, MBPP 84.82. Instruction following IFEval 80.22, hallucination resistance TruthfulQA 76.38.
278K ContextHierarchical Memory Architecture (HBM) preserved: STM (4K, AES-256-XTS) + ITM (64K, AES-256-XTS) + LTM (278K, ChaCha20-Poly1305 + TPM-sealed). Frontier models offer only KV-cache; AIGENCY V4 holds persistent cross-session context 4.3× wider.
Security & ComplianceCompliance with KVKK §5/§12, ISO/IEC 27001, ETSI EN 303 645, NIST SP 800-207 (Zero-Trust), EU AI Act. Post-quantum migration active in Q2 2026: LTM encryption with XChaCha-Kyber1024 hybrid. Multimodal image cache auto-deleted with 30 MB limit + 24h TTL.
WhitepaperFull technical documentation published (32 pages, TR and EN). Architectural specification, training policy, 22 benchmark results, frontier comparison, known limitations, and V4.1/V4.2/V5 roadmap. View on the whitepaper page or download as PDF.
3.0May 30, 2025
Update
NewFully independent model development, domestic and national artificial intelligence with 120B parameters was made available.
2.5.1January 4, 2025
Update
NEW
2.5December 24, 2024
Update
NEWAll models and systems have been updated.
1.05December 1, 2023
Update
NEWNew sub-models have been added.
1.04October 1, 2023
Update
NEWInterface Update
1.0.3March 1, 2023
Update
NEWModel Update
1.0.2.1June 30, 2022
Update
NEWUpdate of Law and Health Models
1.0.2June 1, 2022
Fixed
FİXModel parameter issues have been fixed.
1.0.1May 14, 2022
Model Update
NEWThe model named Alparslan has been activated.