Slack
News
Refix takes on the product, AI, and market shifts that change how teams operate.
Claude Opus 5.5: what it does, what it costs, and when to use it
Claude Opus 5.5 is Anthropic's new flagship, released September 22, 2026. It matches Claude Fable 5.1 on most work and costs about 40% less to run than Opus 5. Here is what that means in practice.
Neil Agarwal
-
Claude Opus 5.5 vs GPT-6 Sol: how the same-day releases compare
Anthropic and OpenAI both shipped flagship models on September 22, 2026. Opus 5.5 is the costlier, agentic-first flagship; GPT-6 Sol is half the list price and built to scale. Here is what the sources actually settle.
Neil Agarwal
-
Claude Opus 5.5 vs Sonnet 5: when the flagship earns its price
Opus 5.5 costs double Sonnet 5 on every price line. Whether that is worth it depends on how much of your work is long-running agentic work, and how you route between the two.
Neil Agarwal
-
GPT-6 Sol: what it does, what it costs, and where it fits
GPT-6 Sol is OpenAI's model for complex coding and agentic workflows, released September 22, 2026 alongside GPT-6 Luna. It carries much of GPT-6 Astra's capability at a fifth of Astra's price.
Neil Agarwal
-
GPT-6 Sol vs GPT-6 Luna: choosing inside the GPT-6 family
GPT-6 Sol and GPT-6 Luna shipped together and share the same context window and tools. Sol costs 20 times Luna's input price, so the choice comes down to workload mix and routing.
Neil Agarwal
-
Jev Choice, Score, and Noul: which primitive should you use?
Choice selects an allowed option, Score evaluates an ordered rubric, and Noul returns the probability of yes. The hard part is matching the primitive to the decision.
Neil Agarwal
-
Jev confidence in production: thresholds, review, and model pinning
Jev confidence summarizes the shape of a Choice or Score distribution. Production systems still need risk-based thresholds, labeled evaluations, fallbacks, and version control.
Neil Agarwal
-
Jev for AI agents: routing, guardrails, and tool selection
Jev can sit inside an agent harness as a fast decision layer for model routing, tool selection, prompt-injection checks, output review, and human escalation.
Neil Agarwal
-
Jev for marketers: classification, QA, and content operations
Jev does not write campaign copy. It can classify intent, score relevance, check supplied evidence, route feedback, and add review gates to repeatable marketing workflows.
Neil Agarwal
-
Jev for product managers: finding the right decision workflow
Product managers do not need to start with the model. Start with a repeated judgment, define the allowed outcomes, price the mistakes, and design the review path.
Neil Agarwal
-
Jev pricing, latency, and benchmarks: how to read the claims
TypeSafe publishes unusually low input pricing and large workflow gains for Jev. The useful question is what those first-party numbers measure and how to reproduce them on your workload.
Neil Agarwal
-
Jev Python tutorial: classify and score requests
Use TypeSafe's synchronous or asynchronous Python client to classify requests, score urgency, inspect probabilities, and route uncertain cases for review.
Neil Agarwal
-
Jev TypeScript quickstart: build a typed ticket router
Install the TypeSafe JavaScript SDK, ask Jev three typed questions, and route a support ticket with confidence-aware application code.
Neil Agarwal
-
Jev for RAG: reranking, passage filters, and citation checks
Jev can score retrieved passages, filter prompt injection and contradictions, rerank candidates, and test whether a source supports a generated claim.
Neil Agarwal
-
What is Jev? TypeSafe AI's System One model explained
Jev is an AI model for bounded software decisions rather than generated text. It returns typed choices, scores, and yes-or-no probabilities that code can use directly.
Neil Agarwal
-
Jev vs. LLMs: typed decisions or generated text?
Jev and conventional LLMs solve different interface problems. Jev returns bounded probabilities for code, while LLMs generate language, code, and explanations.
Neil Agarwal
-
Muse AI launches: Meta's personal agent can browse, plan, and act
Muse is Meta's new personal AI agent. It can work on a goal in the background, use a browser, build tools, and pause for approval before sensitive actions.
Refix
-
Muse vs. Claude vs. Manus: where Meta's personal agent fits
Muse, Claude, and Manus all promise capable AI. Their public positioning points at different jobs: personal coordination, problem solving, and task execution.
Refix
-
Muse vs. Grok Bot vs. Instinct: the new personal agent race
Muse, Grok Bot, and Instinct all point beyond chat toward assistants and agents that can use tools. Their public materials reveal three different bets on where that assistant belongs.
Refix
-
What can Muse AI do? The features behind Meta's personal agent
Muse is built to run longer tasks, browse the web, use connected apps, make artifacts, and ask before high-stakes actions. Here is how each part works.
Refix
Put a real owner on the next decision.
Refix is an AI growth agent. It investigates the signals around your goal and brings your team a next step with evidence to back it up.