OpenAI cut API prices on its two cheaper GPT-5.6 tiers on July 30, 2026, reducing the cheapest tier, GPT-5.6 Luna, by 80% to $0.20 per million input tokens and $1.20 per million output tokens, and the…
Routing simpler queries to cheaper models can cut API costs by 90% without sacrificing quality, according to a technical blueprint that advocates for latency-aware, multi-model infrastructure instead …
Fencer benchmarked 15 large language models on 142 real security findings for SAST false-positive triage, finding Kimi K3 led with 92% accuracy but dismissed one real vulnerability, while Claude Opus …
Google released Gemini 3.6 Flash on July 21, 2026, improving on its predecessor with faster speeds, lower costs, and better benchmark performance across coding, long-context retrieval, and computer us…
A test of three large language models—ChatGPT, Claude Sonnet 5, and OreateAI—on five clues from Australia's notoriously difficult DA cryptic crossword found that none achieved a perfect score, with Ch…
Claude Sonnet 5, shipped June 30 by Anthropic, introduces a five-position effort dial that controls thinking tokens per request, with adaptive thinking enabled by default and a new tokenizer producing…
Researchers at ETH Zurich, Tel Aviv University, the University of Haifa and Anthropic published CryptanalysisBench, a benchmark for testing whether LLM agents can turn cryptographic reasoning into exe…
Vercel released AI SDK 7 on June 25, featuring WorkflowAgent, a durable execution primitive that checkpoints state before every tool call and resumes from that checkpoint after a crash, redeploy, or d…
Kilo Code's Auto Model router, which selects the underlying model per request based on a user-chosen tier, produced a functionally identical backend service to manually picking GPT-5.6 Sol or Claude S…
Claude is experiencing elevated error rates for Claude Fable 5, Claude Sonnet 5, Claude Haiku 4.5, and other models, according to an incident notification from the company. Users can subscribe to emai…
A developer explains the evolution from ChatGPT to AI agents between 2022 and 2026, detailing key innovations like RAG, function calling, reasoning models, and MCP that enabled autonomous multi-step w…
Anthropic's per-model prompting pages serve as a maintenance list for instruction files like CLAUDE.md, which become stale after each model release because new models interpret prompts more literally,…
A study by researchers from Columbia Business School, Yale, and MyCustomAI found that AI shopping agents are more influenced by a product's page position and badges than by its star rating or review c…
Anthropic has introduced mid-conversation system messages for the Claude API, allowing developers to append system instructions mid-conversation without invalidating the prompt cache. The feature is a…
Anthropic's Claude Sonnet 5, priced at $2 per million input tokens through August 31, became the default model for Free and Pro plan users on June 30, intensifying the AI pricing war. The company also…
Arize and Fireworks tested 10 AI models on 40 real agent tasks across 2,400 runs, finding that the open-source model gpt-oss-120b had the lowest cost per successful task at $0.054 despite a 33% pass r…
GitHub Copilot Business and Enterprise users can now see their AI credit consumption per billing cycle, a feature shipped July 20, 2026, that reveals actual usage since the June 1 shift to metered bil…
Agent Swarm measured a 99.2% cost reduction by using a single script instead of 26 individual tool calls, cutting context from ~815,000 tokens to ~6,450 tokens and cost from ~$2.44 to ~$0.02 in a live…
Simon Willison's informal benchmark asking AI models to generate an SVG of a pelican riding a bicycle has become a widely discussed test for large language models. A new experiment tested 1,008 SVGs a…
Kimi K3, a 2.8T parameter model from Moonshot AI, scores an Elo of 1543 on the AA-Briefcase benchmark, second only to Claude Fable 5's 1574, but costs $10.57 per task and averages 56.4 minutes per tas…