Trust, Evaluation & Reliability — Explore AI Tools & Stories

Tools for this

Browse all ->

Artificial Analysis

Independent AI model and API provider benchmarking

LangSmith

The platform for agent engineering.

Vals AI

Benchmark Generative AI for Enterprise Applications.

Claude Security

From scan to fix, done seamlessly

Daytona

Secure, elastic infrastructure for AI agents.

AA-Briefcase

Agentic Knowledge Work Benchmark

AI Gateway

One API for all AI models.

Code Arena

Build & Test with AI Coding Models

Fresh stories

New

Cua launches Spaces for agent-controlled desktops

Cua Spaces provides agent-controlled desktops on local or remote machines. The free, source-available release includes approved app-session transfers, a locked local Keyvault, and separate human and agent cursors.

Release⚙️Agent runtime infrastructure2nd October

Breaking

Meta reports Muse Spark solutions to six open mathematics problems

Meta published six research collaborations reporting solutions to open mathematics problems with Muse Spark. Mathematicians guided the work through its regular chat interface, independently reviewed results, and marked human contributions.

New

Muse·2nd October·5 min read

New

NerfBench says Claude Opus 5.5's 94.2% score is within normal variance

NerfBench reports Claude Opus 5.5 at 94.2% of its launch baseline but says the drop remains within normal variance. Theo disputes the degradation framing and offers to fund independently audited testing.

🧠Claude2nd October

New2nd October

Briefs forOctober 2

Top storiesthis week

See all →

Breaking

Reports rank Gemini 4 Argon highly on four engineering benchmarks

Reports place Gemini 4 Argon at 77.9% on DeepSWE, 57.6% on Terminal-Bench, 77.5% on AutomationBench-AA, and 68% on CWE-Bench. The reports also cite lower cost or token use than rivals.

New

Gemini·30th September·6 min read

New

Google rolls out Gemini 4 Argon to cyber defenders

Google is rolling out Gemini 4 Argon through Fairwind to government users, vetted cyber defenders, and trusted testers. The model supports up to 1M output tokens and costs $2/M input and $10/M output.

Release🧠Gemini30th September

GPT-6.1 Sol scores 86.3% on MathArena BrokenArXiv

GPT-6.1 Sol scored 86.3% on MathArena BrokenArXiv, 78.6% on Braintrust problem solving, and third in Code Arena WebDev. A user also reports that it outperformed its mini-SWE result in Codex.

🧠GPT-6.1 Sol30th September

New

Perplexity open-sources pplx-embed-v2-context-9b-preview

Perplexity released pplx-embed-v2-context-9b-preview, which encodes chunks using whole-document context. Perplexity reports leading results on ConTEB and Turbopuffer's context benchmark.

Release🔎Context Engineering30th September

New

Developers report Firebase payload crashed iOS apps

Developers traced widespread iOS app crashes to a malformed server-side Firebase payload, which was rolled back. Cached payloads reportedly kept some apps failing afterward.

⌨️Reliability29th September

New

Anthropic reports GLM-5.3 built browser exploits in 50 of 410 attempts

Anthropic reports that GLM-5.3 produced working browser exploits in 50 of 410 controlled attempts. In a separate binary-exploitation test, it achieved control-flow hijacks in 4% of trials.

🧠GLM29th September

New

OpenAI releases GPT-6.1 Sol at $2 per million input tokens

GPT-6.1 Sol is available in the API, Codex and ChatGPT Work. It costs $2 per million input tokens and $10 per million output tokens; OpenAI claims near-Astra coding results.

Release🧠GPT-6.1 Sol29th September

New

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.

Release🧠Computer Use28th September

New

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.

Release⚙️Agent runtime infrastructure28th September

New

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.

🧠Claude28th September

See all stories →

New

Reports rank Gemini 4 Argon highly on four engineering benchmarks

Reports place Gemini 4 Argon at 77.9% on DeepSWE, 57.6% on Terminal-Bench, 77.5% on AutomationBench-AA, and 68% on CWE-Bench. The reports also cite lower cost or token use than rivals.

🧠GeminiCoding Agents30th September · 6 min read

Google rolls out Gemini 4 Argon to cyber defenders

Release🧠Gemini30th September

GPT-6.1 Sol scores 86.3% on MathArena BrokenArXiv

GPT-6.1 Sol scored 86.3% on MathArena BrokenArXiv, 78.6% on Braintrust problem solving, and third in Code Arena WebDev. A user also reports that it outperformed its mini-SWE result in Codex.

🧠GPT-6.1 Sol30th September

Perplexity open-sources pplx-embed-v2-context-9b-preview

Perplexity released pplx-embed-v2-context-9b-preview, which encodes chunks using whole-document context. Perplexity reports leading results on ConTEB and Turbopuffer's context benchmark.

Release🔎Context Engineering30th September

Developers report Firebase payload crashed iOS apps

Developers traced widespread iOS app crashes to a malformed server-side Firebase payload, which was rolled back. Cached payloads reportedly kept some apps failing afterward.

⌨️Reliability29th September

Anthropic reports GLM-5.3 built browser exploits in 50 of 410 attempts

🧠GLM29th September

OpenAI releases GPT-6.1 Sol at $2 per million input tokens

Release🧠GPT-6.1 Sol29th September

H Company releases 27B and 35B-A3B Holo4 computer-use models

Release🧠Computer Use28th September

NVIDIA launches OpenShell to isolate AI agents

Release⚙️Agent runtime infrastructure28th September

Independent tests put Sonnet 5.5 at $7.60 per max-effort task

🧠Claude28th September

Daily AI Digest

Get the best stories delivered
to your inbox

Explore what's new in AI

Filters

Tools for this

Fresh stories

Cua launches Spaces for agent-controlled desktops

Meta reports Muse Spark solutions to six open mathematics problems

NerfBench says Claude Opus 5.5's 94.2% score is within normal variance

Cua launches Spaces for agent-controlled desktops

Meta reports Muse Spark solutions to six open mathematics problems

NerfBench says Claude Opus 5.5's 94.2% score is within normal variance

Briefs forOctober 2

Top storiesthis week

Reports rank Gemini 4 Argon highly on four engineering benchmarks

Google rolls out Gemini 4 Argon to cyber defenders

GPT-6.1 Sol scores 86.3% on MathArena BrokenArXiv

Perplexity open-sources pplx-embed-v2-context-9b-preview

Developers report Firebase payload crashed iOS apps

Anthropic reports GLM-5.3 built browser exploits in 50 of 410 attempts

OpenAI releases GPT-6.1 Sol at $2 per million input tokens

H Company releases 27B and 35B-A3B Holo4 computer-use models

NVIDIA launches OpenShell to isolate AI agents

Independent tests put Sonnet 5.5 at $7.60 per max-effort task

Reports rank Gemini 4 Argon highly on four engineering benchmarks

Google rolls out Gemini 4 Argon to cyber defenders

GPT-6.1 Sol scores 86.3% on MathArena BrokenArXiv

Perplexity open-sources pplx-embed-v2-context-9b-preview

Developers report Firebase payload crashed iOS apps

Anthropic reports GLM-5.3 built browser exploits in 50 of 410 attempts

OpenAI releases GPT-6.1 Sol at $2 per million input tokens

H Company releases 27B and 35B-A3B Holo4 computer-use models

NVIDIA launches OpenShell to isolate AI agents

Independent tests put Sonnet 5.5 at $7.60 per max-effort task

Daily AI Digest