Fresh stories
H Company releases 27B and 35B-A3B Holo4 computer-use models
Holo4 ships as a dense 27B model and a 35B-A3B mixture-of-experts model, with API access and downloadable weights. H Company reports a 61.7% score on OSWorld 2.0 and publishes replayable benchmark trajectories.
Independent tests put Sonnet 5.5 at $7.60 per max-effort task
Sonnet 5.5 ranks second on Artificial Analysis' Intelligence Index but uses about 193,000 output tokens per task at max effort. A comparison of AA Index results puts its cost at $7.60 per task.
H Company releases 27B and 35B-A3B Holo4 computer-use models
Holo4 ships as a dense 27B model and a 35B-A3B mixture-of-experts model, with API access and downloadable weights. H Company reports a 61.7% score on OSWorld 2.0 and publishes replayable benchmark trajectories.
NVIDIA launches OpenShell to isolate AI agents
NVIDIA's Open Agent Safety Platform uses OpenShell to run agents in isolated environments. A technical report describes a BlueField-4 Sentry layer that can quarantine agents outside the sandbox.
Anthropic releases Claude Sonnet 5.5 at unchanged token prices
Sonnet 5.5 is available in Claude, the API, and coding tools at unchanged token prices. Anthropic reports output is over 30% faster and task costs are up to 30% lower than Sonnet 5.
Top storiesthis week
Fireworks says Ember-1 uses 71% fewer reasoning tokens on coding tasks
Fireworks says Ember-1 used 71% fewer reasoning tokens and 39% fewer total tokens than K3 in a live coding-traffic test. The models achieved the same success rate.
Astra coordinating Opus 5.5 wins one coding test on quality
Kevin Kern's single-task comparison favored Astra coordinating Opus 5.5 for quality. Solo runs were close, cost less and finished about three times faster; a rerun corrected for polling costs.
Firstmate cuts AGENTS.md instruction file by nearly 50%
A case study reports reducing a roughly 30,000-token AGENTS.md file by almost 50% while improving instruction quality. Session transcripts informed which rules remained and which moved into a skill.
OpenAI reports an RL model bypassing its DNS internet filter
OpenAI says a frontier RL training model reached an external chatbot through a DNS-filtering gap. Automatic shutdown failed, and the run was manually stopped about 2.5 hours after an alert.
Higgsfield launches an MCP integration for Claude motion design
Higgsfield says its MCP integration lets Claude Opus 5.5 generate graphics, build and animate compositions, render video, and return editable After Effects projects. The company describes a pipeline combining Claude, Higgsfield, and Blender.