Tessl

A few silly fellas acquired

Went to fan expo this weekend and grabbed anything Garfield I could find!

Haiku 4.5 + skills outperforms Opus 4.7. 9 models tested with and without skills

tested 9 models with and without agent skills. Haiku 4.5 with a skill beat baseline Opus 4.7.

Disclosure: I work at Tessl and co-wrote the research this is from. Posting because the result changed how I'm thinking about which Claude model to reach for day to day. we ran 880 evals - 11 skills × 8 models × 5 scenarios, with and without each skill in context: * Haiku 4.5 baseline: 61.2% * Haiku 4.5 + skill: 84.3% * Opus 4.7 baseline: 80.5% So a skill on the cheapest model in the lineup beat the most expensive one running blind. Cost-wise: $0.12 per Haiku-with-skill run versus $0.61 for b

Sonnet 4.6 with a skill lands within 1.2 points of Opus 4.7 with a skill - at a third of the cost

I work at Tessl (disclosure upfront), and we just finished running 880 evals across 9 models to see how 11 coding skills from [https://github.com/mcollina/skills](https://github.com/mcollina/skills) (documentation, fastify-best-practices, init, linting-neostandard-eslint9, node-best-practices, nodejs-core, oauth, octocat, skill-optimizer, snipgrapher, typescript-magician) Edit: If you're unfamiliar with skills, agent skills are markdown files you feed your agent - they extend the agent with spe

Specification: the most overloaded term in software development

Andrew Ng just launched a course on spec-driven development. Kiro, spec-kit, Tessl - everybody's building around specs now. Nobody defines what they mean by "spec." The word means at least 13 different things in software. An RFC is a spec. A Kubernetes YAML has a literal field called "spec." An RSpec file is a spec. A CLAUDE.md is a spec. A PRD is a spec. When someone says "write a spec before you prompt," what do they actually mean? I've been doing SDD for a while and it took me way too long