Insights, tutorials, and updates from the ZharfAI team
Engineering, infrastructure, and emerging AI systems
Haiku 5.5 brings a million-token context and much lower prices. The 100K pricing threshold, effort settings, and API changes determine what teams actually save.
Reading Anthropic's 144-page Haiku 5.5 system card: benchmark conditions, prompt injection, sandbox behavior, honesty regressions, oversight, and model welfare.
Sonnet 5.5 cuts cache-read pricing on October 7. We examine task costs, code-review evidence, API changes, and when Opus 5.5 still earns its higher price.
A guide to OpenAI's July–September 2026 releases: GPT-6 Sol and Luna, Astra, image and voice models, with API prices, access limits, and migration choices.
How Iranian teams can reach LLM APIs lawfully: domestic providers, gateways, self-hosted open models, Persian evaluation, a cost method, and a client that swaps providers by config.
Anthropic's September 22 model leads most agentic benchmarks, cuts cache reads by 60%, and brings four breaking API changes and Fable-class safeguards to Opus.
A section-by-section reading of Anthropic's 230-page Opus 5.5 system card: CB and AI R&D thresholds, cyber evaluations, prompt injection, alignment audits, welfare, and benchmarks.
Jev returns typed answers with calibrated probabilities instead of prose. What TypeSafe's first System One model does, what its launch evidence shows, and where it falls short.
DeepSeek's September 10 release adds native vision and cheaper agent workloads. We examine the benchmarks, API prices, and September 14 migration deadline.
A technical reading of DeepSeek's 51-page report: causal encoder–decoder execution, CSA2, bounded replay, Engram, DSpark, and what the reference code omits.
OpenAI's Astra launch brings stronger scientific and computer work, new agent APIs, and premium pricing. Here is the evidence and a practical adoption guide.
Anthropic's September 1 release lifts agentic coding and scientific work, cuts cache-read pricing by 75%, and changes how production harnesses preserve reasoning.