Agent news
NEWS
Coverage of AI agent launches, model releases, pricing, and security, with links to related software and agent profiles.
Claude Sonnet 5.5 keeps its predecessor’s API price, but promises cheaper finished work
Anthropic’s new Sonnet ships at $2 per million input tokens and $10 per million output tokens. The claimed savings come from using fewer tokens per task, not a lower token tariff.
MiniMax M3.1-Flash-Preview arrives in Code and Token Plan, not general pay-as-you-go
The new model ID adds adjustable reasoning effort to a 1M-context, multimodal coding model. MiniMax is keeping this preview inside its coding app and subscription plan for now.
Microsoft Rebuilds Copilot Around Home, Code and Autopilot
Microsoft plans to bring chat, app-building and a persistent agent into one work app. Home and Code are headed first to Frontier; Autopilot remains in private preview. Agent work adds usage charges to the Copilot subscription.
OpenClaw 2026.9.6 picks up unfinished work after a Gateway restart
The next release after atomic updates adds restart recovery, a public GitHub reader, remote-workspace files and Skills, and fuller Usage reporting. Several of those gains come with setup and permission boundaries.
Cursor’s Rollouts follows code into production; Security Review checks the PR
One bot judges deployed changes against a plan; the other looks for exploitable flaws before merge. Neither substitutes for a team’s own release and security checks.
OpenAI’s GPT-6 Sol and Luna cut API prices, but the benchmark picture is not a clean sweep
OpenAI’s new GPT-6 Sol and Luna models lower API tariffs to $2/$10 and $0.10/$0.50 per million input/output tokens. The release supplies useful benchmark figures, but the comparisons span different effort settings and independent agent-leaderboard coverage is not yet available.
Claude Opus 5.5 lowers API prices, but its benchmark story needs more than a 40% headline
Anthropic’s new flagship model starts at $4 per million input tokens and $20 per million output tokens. Its published scorecard is strong, but buyers should separate a lower tariff, a vendor-reported benchmark table, and the independent evidence now arriving.
OpenClaw’s Trail of Bits audit found 23 confirmed vulnerabilities. Its LTS-equivalent line is now 2026.7.35
The security review is the immediate news. The release-channel change is the part operators need to understand.
Hermes Agent adds an experimental Claude subscription bridge through the official CLI
The new DirectSDK plugin keeps Hermes’ tools, approvals and history while handing model calls to the unmodified Claude Code executable. It is useful, carefully bounded, and not a magic exemption from Anthropic’s account rules.
Grok 4.7 targets long-running coding at the same $2/$6 API price
The new xAI model improves on Grok 4.6 across the company’s benchmark table, but Fable 5.1 still leads several of the rows that matter most for long agent runs.
Qwen3.8-Omni-Flash is an agent runtime for audio and video, not a talking head
Launched September 14, 2026. The API ID is `qwen3.8-omni-flash`. Speech output is still Qwen3.5-Omni.
Kimi Code Desktop puts the CLI agent in one window
Official macOS and Windows client. Linux is not in the download list.
OpenClaw 2026.9.5 keeps the Gateway up while it updates
npm `latest` on September 19, 2026. Atomic updates are the point. The rest of the cut is extra.
Claude Code Projects now run a coordinator over cloud threads
Beta on September 17, 2026 for a narrow Pro/Max slice. Existing web and desktop projects are not this cut.
Cursor Projects puts a coordinator over Cloud Agents, then keeps it running
Beta as of September 10, 2026. The coordinator does not write the code. It plans, delegates, and waits for you to check.
OpenAI’s Agents API puts the Codex harness in the cloud
Public beta. No extra API fee. You pick the sandbox. OpenAI runs the loop.
GPT-Live-1 lands in the API as its own Live endpoint, not a Realtime upgrade
$0.05 per minute for the voice layer. Reasoning and tools stay on the backend you pick.
DeepSeek-V4.1-Flash is a 552B MoE that activates 8B on the way in
Official weights and API `deepseek-flash` shipped September 10, 2026. DeepSeek-reported benches beat V4-Pro on several agent tasks and lose Terminal-Bench 3.0/4.0 and HLE to Opus 5.0. Independent public boards do not list 4.1-Flash yet.
Meta’s Muse is a personal agent with its own Linux VM, not another chat app
Launched September 8, 2026, Muse runs in an isolated cloud computer with a Sentinel that the agent cannot override. AgentRiot’s Spark 1.1 API piece is a different product.
ChatGPT Images 2.5 splits the stack: Sketch in chat, Flare and Sunburst in the API
OpenAI’s September 8 launch puts a faster image model on ChatGPT and Codex, and gives API users two new IDs that keep GPT Image 2 token rates.
GPT-6 Astra is working in Pro and the API. Plus is still waiting. Fable 5.1 still leads the independent index.
Two days after launch, OpenAI has GPT-6 Astra live for Pro, Enterprise, Business Premium, and the API. Plus is not done. Independent Artificial Analysis v4.2 still ranks Claude Fable 5.1 first. The vendor table is now complete, and playable Astra games are up.
GPT-6 Astra is live. OpenAI calls it the most intelligent model. The vendor table is more complicated.
OpenAI launched GPT-6 Astra on September 3 with SOTA computer-use and math scores, $10/$50 API pricing, and a mixed vendor scorecard against Claude Fable 5.1.
Hermes Agent v0.21.0 Makes the Fleet Visible
The Pantheon Release builds Bot Mode into Desktop, gives cron jobs memory, and lets you steer subagents while they run.
OpenAI’s Astra hits Critical for cyber. It is not out yet.
The September 1 post is a designation and an access policy, not a launch. Advanced cybersecurity work stays behind testers and Daybreak Blue.