multi-agent

A paper from Meta FAIR and Harvard University researchers, published October 12, 2026, proposes ScaleMAS, a coordination mechanism for multi-agent AI systems. On a software-fix task, four ScaleMAS agents matched eight independent agents using nearly half the tokens, while gains from adding agents varied sharply by task type.

Researchers from Sakana AI and UC Berkeley introduced Multi-Agent Self-Supervision (MASS), in which a single model solves tasks, judges results and optimizes workflows. Applied to Qwen3.6-27B, the win rate against the base model reached 69.9% after two rounds. But the self-evaluator sometimes accepted flawed or manipulated solutions as successes and fed them into the next training cycle.

Vals AI tested GPT-6 Sol and Claude Opus 5.5 on Vibe Code Bench as solo agents and as teams. Teams cost 1.8x to 5.1x more, and only one of four comparisons showed a statistically significant improvement. Anthropic's own scaling data and comments from OpenAI's Noam Brown point the same way: more agents mostly buy speed, not quality.

Anthropic has added dynamic workflows to Claude Managed Agents, letting a lead agent plan a task, distribute it to up to 1,000 parallel sub-agents per execution, and merge the results. Anthropic claims the approach found 66 of 70 hidden bugs in a 116,000-line codebase, versus 14 to 27 for a single agent. Pricing and token costs were not disclosed.

Sakana AI has released Fugu Ultra, a multi-agent orchestration system that routes tasks across pools of underlying models rather than operating as a single monolithic model. The system supports a 1M token context window and is priced at $5 per million input tokens and $30 per million output tokens.

Sakana AI released Fugu, an orchestration language model that routes tasks across multiple AI providers to reduce vendor lock-in risks. The Japanese AI firm positions Fugu as a solution to enterprise dependency on single monolithic AI APIs.

Augment has released Cosmos, a platform designed to coordinate multiple AI agents across the entire software development lifecycle. The system enables agents to work on issue triage, code review, build fixes, and incident response while integrating with tools like Slack and Linear.

Anthropic released Claude Opus 4.8 on Thursday, its fastest upgrade cycle at 41 days since the previous Opus 4.7. The model includes a new Dynamic Workflows feature designed to manage complex tasks across hundreds of parallel subagents, with pricing unchanged from previous Opus releases.

xAI has released Grok 4.20 Multi-Agent, a variant designed for collaborative agent-based workflows with a 2-million-token context window. The model scales from 4 agents at low/medium reasoning effort to 16 agents at high/xhigh effort levels, priced at $2 per million input tokens and $6 per million output tokens.

Anthropic has launched Code Review, a multi-agent system within Claude Code that automatically analyzes AI-generated code and flags logic errors. The tool addresses enterprise concerns about managing the increasing volume of code produced by AI systems.

Samsung is expanding Galaxy AI to include Perplexity as an integrated AI agent accessible via the voice command "Hey, Plex." The integration gives Perplexity access to Samsung's native apps including Notes, Clock, Gallery, Reminder, and Calendar, marking Samsung's shift toward a multi-agent ecosystem.