Tag: Technology & Innovation

Algorand Aims for Quantum Resistance by 2027

The Proof-of-Stake (PoS) blockchain network, Algorand, has announced its intention to achieve quantum resistance by the end of 2027, a move that comes amid growing concerns about the potential threat of quantum computing to cryptocurrency security.

In a recent blog post, the Algorand team highlighted the significant risk that quantum computing poses to the security of blockchain technology. To address this, the network has laid out a comprehensive roadmap detailing its research, plans, progress, and milestones in the field of post-quantum cryptography (PQC).

Algorand has been recognized by Google Quantum AI as one of the smart-contract platforms capable of implementing PQC. The network has already conducted its first PQC-secured transaction in 2025 and aims to complete its full transition to PQC within the next two years.

“Post-quantum migration is a balancing act,” the Algorand team stated. “Moving too slowly leaves systems exposed to future quantum attacks, but moving too quickly can mean relying on algorithms and implementations that have not yet been sufficiently battle-tested.”

The initial phase of Algorand’s roadmap involves introducing native post-quantum account support in a protocol release scheduled for the third quarter of 2026. While Falcon accounts, previously enabled via the Algorand Virtual Machine (AVM), have demonstrated the feasibility of post-quantum signatures on the protocol, they are not yet natively supported by the ledger. The forthcoming native support will enable network-level compatibility with multiple concurrent signature schemes.

Following the standardization of a new derivation scheme for lattice-based post-quantum keys, Algorand plans to integrate PQC updates across its suite of tools, including legacy software development kits (SDKs), hardware wallets, and AlgoKit. The network will eventually extend support to additional signature schemes for traditional Ed25519 accounts.

Throughout these upgrades, Algorand is committed to fostering an environment that allows for the seamless integration of future advancements with minimal disruption to the protocol.

“Building on our robust history of native multisig, the arrival of cryptographic agility and native post-quantum accounts enables us to deploy native multisig support for multi-cryptography schemes by the end of 2026,” the team explained. “We view this as an essential advancement for institutional operations, treasury management, and high-stakes financial applications.”

A subsequent step in the roadmap will explore post-quantum multisignatures as a generic policy layer, enabling weighted approvals, hybrid combinations of classical and post-quantum signers, and the incorporation of future PQC signature algorithms as they become standardized. This measure is designed to provide robust protection against both current and future quantum-era threats.

Algorand is not alone in its pursuit of quantum resistance; other blockchain networks, including Ethereum and Ripple, are also actively working towards this critical milestone.

OpenRouter has launched an API built on a straightforward premise: that a carefully orchestrated ensemble of less expensive AI models can rival the performance of a single, high-cost one. The benchmark for this “expensive” model is Anthropic’s Claude Fable 5.

The new product, named Fusion, operates by sending a user’s prompt to multiple AI models simultaneously. A designated “judge” model then analyzes these parallel responses to identify consensus points, discrepancies, and omissions. Finally, a “synthesizer” model compiles this analysis into a single, coherent, and well-supported answer.

This launch arrives at a particularly opportune moment. Just last week, following the release of Fable 5 and Mythos 5, a U.S. export control directive compelled Anthropic to suspend access to these models for all non-U.S. individuals globally, citing a contentious jailbreak finding. OpenRouter seized on this development, posting to X the following day and directly addressing the resulting void with a promise of “Fable-level intelligence at half the price.”

Introducing Fusion

Fusion, the company’s new offering, is positioned as the most intelligent compound model currently available on the market, capable of delivering Fable-level intelligence at a significantly reduced cost.

How Fusion Works

When a prompt is submitted to Fusion, OpenRouter dispatches it concurrently to a selected group of AI models. Each model in this panel is equipped with web search and bash tool capabilities.

Following the initial response generation, a judge model scrutinizes each output, pinpointing areas of agreement, identifying contradictions, and noting any gaps in the information. Once this analytical phase is complete, a synthesizer—by default, Claude Opus 4.8—constructs the final answer, ensuring it is grounded in the preceding analysis.

The entire process is managed server-side. Users can opt for a default model panel by setting their model string to “openrouter/fusion.” Alternatively, they can integrate Fusion as a tool within their own model calls for selective use, or even construct custom panels through a no-code interface in the Fusion chatroom.

Performance Benchmarks

OpenRouter subjected Fusion to testing on DRACO, a benchmark developed by Perplexity that utilizes real-world user research requests. In these tests, a panel comprising Fable 5, OpenAI’s GPT-5.5, and synthesized by Opus achieved a top score of 69%. A solo Fable model scored 65.3%, though it failed to complete 10% of its tasks due to its own content filters.

Of particular note is the performance of a more economical combination: Gemini 3 Flash, the open-source Chinese models Kimi K2.6 and DeepSeek V4 Pro, all fused and synthesized by Opus. This ensemble reached 64.7%, outperforming solo GPT-5.5 (60%) and solo Opus 4.8 (58.8%) and coming within one percentage point of Fable’s score, all at approximately half the cost.

Even a configuration pairing Opus 4.8 with a separate instance of itself yielded a score of 65.5%, a 6.7-point improvement over solo Opus. OpenRouter attributes roughly three-quarters of this gain to the synthesis process itself, with the remainder stemming from genuine model diversity.

One unexpected issue arose when granting the panel live web access: models surfaced DRACO’s own grading rubric within search results, a potential contamination risk that OpenRouter described as coincidental rather than intentional. The issue was resolved with a single configuration change to exclude the benchmark’s hosting domains from the search tools, and all published figures reflect this corrected run.

OpenRouter acknowledges that Fusion is not a complete substitute for Fable. DRACO’s benchmark excludes long-horizon tasks, an area where Fable is reportedly still superior. For coding tasks, Fusion functions as a tool that a coding model can call selectively, rather than a wholesale replacement. This caveat mirrors findings from Decrypt’s testing of DeepClaude, a cheaper backend alternative that maintains Claude Code’s agent loop but still lags behind Opus on the most demanding reasoning tasks.

For everyday operations, standard models remain sufficient. Fusion is designed for queries where a single model might overlook crucial details, and where multiple perspectives can provide a more comprehensive and accurate outcome.

The platform appears particularly beneficial for in-depth research, complex planning, or any scenario where identifying and resolving contradictions is paramount.

The accompanying charts clearly illustrate a key takeaway: for these types of tasks, relying solely on an expensive, single model is no longer the only path to achieving robust synthesis. A collective of readily accessible models, when fused together, can now deliver comparable results at a substantially lower price point.

Reception and Limitations

Sentiment analysis of the launch thread indicated a roughly two-to-one positive reception. AI researcher Andrew Trask hailed the development as “a way bigger deal than it seems,” suggesting that leading AI labs will no longer hold a monopoly on cutting-edge capabilities. However, some skeptics raised concerns regarding the framing, citing poor performance in coding tasks, inadequate tool utilization, and a lack of transparency, particularly given that Fable 5 is no longer accessible for direct comparison.

Fusion operates entirely on models routed through OpenRouter’s own infrastructure, meaning it does not resolve the underlying export control issues affecting models like Fable 5. For users impacted by these restrictions, options now include Fusion panels, backend swaps like DeepClaude, or open-weight alternatives such as GLM-5.2, which may not surpass Fable but offer a viable and cost-effective solution.

Z.AI Unveils GLM-5.2, Challenging AI Dominance with Open-Source Power

Z.AI, a Beijing-based artificial intelligence lab, has launched its latest model, GLM-5.2, on June 16, touting top-tier performance that surpasses its predecessor, GLM 5.1. The release comes as the company, which has been on the U.S. Entity List since January 2025, appears to be capitalizing on growing concerns regarding America’s approach to artificial intelligence. The past week has seen Z.AI’s stock surge by 90%, reaching a new all-time high, fueled by the ban on Anthropic’s Fable and the introduction of GLM-5.2.

GLM-5.2 Demonstrates Strong Benchmark Performance

GLM-5.2’s performance metrics support the considerable hype surrounding its release. On the FrontierSWE benchmark, which assesses an AI agent’s ability to complete open-ended technical projects, GLM-5.2 achieved a dominance rate of 74.4. This places it closely behind Claude Opus 4.8’s 75.1 and ahead of GPT-5.5’s 72.6. In the SWE-bench Pro test, designed to evaluate the autonomous resolution of real-world GitHub issues, GLM-5.2 scored 62.1, outperforming GPT-5.5’s 58.6 and significantly exceeding its predecessor GLM-5.1’s score of 58.4.

The Artificial Analysis Intelligence Index, which aggregates nine different scores to gauge the overall quality of AI models, has recognized GLM-5.2 as the leading open-source model to date. OpenRouter’s benchmarks further position GLM-5.2 in the same performance tier as the now-banned Claude Fable 5.

Hardware and Cost Innovations

A notable aspect of GLM-5.2’s development is the hardware utilized for its training. The model was trained exclusively on Huawei Ascend chips, with no Nvidia components involved. Emad Mostaque, founder of Stability AI, estimated the total training costs at approximately $25 million, with 80% allocated to post-training efforts. This figure represents a significant cost advantage compared to its industry peers.

As previously reported, Z.AI has been training image models on Huawei’s Ascend Atlas servers without using any American chips. GLM-5.2 builds upon this infrastructure, featuring a 744-billion-parameter mixture-of-experts model with a substantial 1 million-token context window. This is a fivefold increase from GLM-5.1’s 200,000-token limit. The model is released under an MIT license, which prevents government directives from restricting access.

Tokens are defined as the segments of text a model can process and generate, while parameters refer to the internal settings and values that dictate how a model interprets information and formulates responses.

Developer Impact and Accessibility

For developers, the expanded context window represents a significant operational shift. Tasks such as navigating entire code repositories, performing multi-file refactors, and executing long agentic pipelines, which previously required breaking down into smaller segments, can now be handled within single API calls. The API pricing for GLM-5.2 is set at $1.40 per million input tokens and $4.40 per million output tokens, a stark contrast to Claude Opus 4.8’s $5 input and $25 output rates. The Coding Plan is available starting at approximately $18 per month and integrates directly with Claude Code, Cline, Kilo Code, and other popular agentic environments.

Local deployment of GLM-5.2 is also technically feasible. Unsloth AI has developed 2-bit GGUF quantizations that reduce the model’s size from 1.51 terabytes to 238 gigabytes while maintaining approximately 82% accuracy. However, this still necessitates a considerable 256 gigabytes of unified memory or a combination of RAM and VRAM, comparable to a fully equipped M4 Ultra Mac Studio or a workstation with a mid-range GPU and 256 gigabytes of system RAM with mixture-of-experts offloading. While substantial, this level of hardware is attainable for users seeking to run the model locally.

Real-World Application and Economic Viability

In a practical test, GLM-5.2 was tasked with building a game that combined typing mechanics with shooter elements. While the user interface was less polished than those generated by other models, the gameplay experience was notably diverse, featuring varied scenarios across waves, shifting enemy types, and later-stage boss encounters. The model generated more varied game states than any other tested for the same task in a zero-shot setup.

The economic advantages of GLM-5.2 are most apparent in workflows requiring output diversity, such as multi-shot generation and agentic pipelines. At open-source pricing levels, the cost-effectiveness is compelling. However, for highly demanding and sustained tasks, such as the SWE-Marathon benchmark where GLM-5.2 scored 13.0 against Opus 4.8’s 26.0, a performance gap with closed-source frontier models remains.

The open-source weights for GLM-5.2 are available on HuggingFace under the MIT license, with quantized weights also accessible. Subscribers to the GLM Coding Plan can now utilize the model by specifying “GLM-5.2” in their model string. Free testing with usage constraints is also available on z.AI.

FinHarbor Unveils Payment Orchestration Module to Streamline Financial Operations

Nicosia, Cyprus – June 18, 2026 – FinHarbor, a provider of modular banking and crypto-acquiring infrastructure, has announced the launch of Money Flow, a new payment orchestration module designed to manage the entire lifecycle of transactions. The module aims to significantly reduce the time it takes for businesses to configure and modify payment processes involving fiat and cryptocurrency, moving from weeks to mere days, while incorporating built-in compliance reporting for every transaction.

Money Flow oversees deposits, withdrawals, transfers, and exchanges across both fiat and crypto networks. At its core, the module utilizes Temporal, a workflow engine engineered for long-running distributed processes. Each payment operation is treated as a stateful process that navigates through Anti-Money Laundering (AML) screening, ledger accounting, and final execution, whether within a traditional bank or on a blockchain. This stateful design ensures that operations can resume precisely where they left off, even if a service restarts mid-process or an external party experiences delays in responding. The system is also equipped to handle situations where a response is not received within a set timeframe, triggering compensating actions or escalating the issue to support, thereby preventing funds from being held in limbo.

The module promises to revolutionize the speed at which payment logic can be adjusted. Modifications that previously required approximately a week can now be implemented in about a day. Furthermore, new payment processes built upon existing integrations can be deployed within days. All flow logic is consolidated within a single service and adheres to a self-documenting code approach, eliminating the need for developers to scrutinize multiple subsystems to understand a specific operation.

This design also bridges the gap between finance leadership and engineering teams. Chief Financial Officers can articulate payment processes in business terms, which developers can then translate into code using a domain-specific language that remains accessible to non-technical stakeholders. This provides finance teams with direct insight into how money moves through the platform, rather than relying on indirect explanations of the underlying logic.

“Payment infrastructure has traditionally been a black box for the people who are ultimately accountable for the money inside it,” said Ilya Podoynitsyn, CEO of FinHarbor. “With Money Flow, a finance director can read the logic of a withdrawal or an exchange almost like a business document, request a change, and see it in production within a day. That changes the conversation between the finance function and engineering.”

Compliance is integrated as a dedicated layer within each workflow. AML rules can be configured by compliance officers or support staff through a visual constructor, depending on the team’s structure. For every payment, the system generates a report that can be shared with regulators or the company’s anti-money-laundering officer, providing licensed businesses with a documented audit trail for crypto-to-fiat operations.

Currently, Money Flow supports bank transfers, crypto withdrawals, payouts, pay-ins, exchanges, crypto and wire deposits, internal transfers, and administrative operations. The range of supported flows is expected to expand as new integrations are added.

“Most of the cost in payment systems comes from change, not from the original build,” Podoynitsyn added. “We designed Money Flow so that change becomes routine – retries, timeouts, and compensations are built into the engine, and teams spend their time on the logic of the business rather than on failure handling.”

The Money Flow module is available to FinHarbor clients as an integral part of the platform’s core infrastructure.

About FinHarbor

FinHarbor is a technical platform provider that enables the launch of compliant, modular financial products, ranging from digital wallets and neobanks to crypto ramps and over-the-counter (OTC) trading desks. Leveraging years of practical fintech experience, the platform encompasses onboarding, compliance, wallet management, transactions, card services, and reporting. It is delivered through a microservice-based architecture (ISO/PCI DSS-certified), a robust API layer, and offers deployment options that are either on-premise or cloud-ready. FinHarbor supports fiat-only, crypto-native, and hybrid business models across markets in Europe, the Middle East and North Africa (MENA), and beyond.

AI Agents Learn to Operate Physical Robots Autonomously

A fleet of eight robotic arms at Nvidia’s GEAR lab has spent the past few weeks independently mastering tasks such as inserting pins, seating graphics cards, and cutting zip ties. The only human involvement was in the subsequent documentation of the process.

This capability stems from ENPIRE, a framework detailed in a paper published Tuesday by researchers from Nvidia, Carnegie Mellon University, and UC Berkeley. ENPIRE delegates the entire robot training process to AI coding agents—the same software that already writes and tests its own code—allowing them to execute this training directly on physical hardware.

For the past year, coding agents like OpenAI’s Codex, Anthropic’s Claude Code, and Moonshot’s Kimi Code have been engaged in what researchers term “autoresearch,” a cycle of writing, testing, and rewriting code without human intervention. This loop has largely remained confined to digital environments, where resetting a failed experiment incurs no cost. ENPIRE, however, extends this process into the physical realm, where resetting an experiment necessitates the manipulation of actual robotic arms.

The system operates in two distinct stages. Initially, a human guides the agent in developing two essential tools: a reset routine to return the workspace to its initial state, and a reward function that analyzes camera footage to evaluate success. This setup, akin to a tireless, ever-vigilant referee, is established once and then reused for all subsequent attempts.

Once these foundational tools are in place, the AI agent assumes complete control. It scours published research for relevant ideas, selects training methodologies such as imitation learning, reinforcement learning, or predefined rules, and then modifies its own code before testing the outcome on the robot. This entire loop requires no human oversight, a development that could be seen as either liberating or slightly disquieting, depending on one’s comfort level with unsupervised robots wielding tools.

Nvidia conducted its experiment using eight bimanual robot stations, each equipped with its own hardware, computer, and coding agent. Progress across these stations is shared via Git, a collaborative coding tool, enabling successful strategies to be disseminated across the entire fleet within minutes.

The effectiveness of the system was measured using two tasks: “Push-T,” where a robot uses only pushes to slide a T-shaped block into a designated zone, and pin insertion, which involves threading pins into 4-millimeter holes. Scaling the operation from one robot to eight reduced the time required to master Push-T from approximately five hours to two, and pin insertion from over 90 minutes to about 40 minutes.

Across the four real-world tasks evaluated, the AI agents achieved a 99% success rate with their learned policies, according to the paper. For pin insertion, the agents reached near-perfect reliability faster than a comparable human-in-the-loop method, which typically requires daily human supervision.

Jim Fan, co-lead of the GEAR Lab and director of Nvidia’s AI research, described the project as the first effort to enable autoresearch in the physical world. Fan stated that the team provided the agents with a fleet of robots, GPU resources, and a token budget, then allowed the robots to autonomously complete the tasks.

Bridging the Simulation-Reality Gap

The disparity between simulated environments and the physical world became apparent almost immediately. While all three coding agents successfully completed the Push-T task within a simulator, two of them failed when the same task was transferred to a physical robot, the paper noted. This highlights the challenges posed by real-world factors such as friction, which are not accurately replicated in simulations.

Nvidia also tested ENPIRE within RoboCasa, a simulated kitchen environment designed to assess robots on chores like opening cabinets and turning off stoves. In this benchmark, ENPIRE surpassed both Nvidia’s own end-to-end model, GR00T, and CaP-X, a tool-using agent that bypasses the autoresearch loop entirely.

ENPIRE builds upon an idea Nvidia first introduced with Eureka in 2023, a system that utilized a language model to generate reward functions for robots within a simulator, eliminating the need for manual engineering. ENPIRE takes this self-improvement loop a step further by moving it from simulation to actual hardware, with the AI agent not only designing its own rewards but also its own testing procedures.

This development coincides with Alibaba’s recent unveiling of its own embodied AI initiative, the Qwen-Robot Suite, a collection of foundation models for robot navigation, manipulation, and physics simulation. While Alibaba is developing software for robots it does not manufacture, Nvidia is exploring the potential for AI agents to manage the entire research and development loop on its own hardware. Both approaches signal a growing trend: physical robots are emerging as the next frontier for AI coding agents to compete and innovate.