Free to readanalytics with consentunsubscribe in one click

The Pack
When a tool ships.

The Pack is how we tell readers something changed. No weekly cadence. Drops mirror what ships in git.

Latest drop

2026-09-05models

September 2026 frontier: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, Hy4 preview

OpenAI's 3 Sep GPT-6 Astra, Google's 2 Sep Gemini 3.8 Flash, and Meta's 2 Sep Muse Spark 1.3 are now on the atlas with live Artificial Analysis Intelligence Index v4.2 scores. Tencent's Hy4 preview is recorded from the official launch. Claude Fable 5.1 still leads the live Vals Index at 68.83%. /models

Read in the atlas →

What actually shipped recently.

Every entry below would have been a line in a drop email. Tool promotions, status changes, privacy re-reads, new playbooks, benchmark updates. No filler entries to hit a word count.

Full ledger · RSS feed →

2026-09-05models

September 2026 frontier: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, Hy4 preview

2026-09-01models

Claude Fable 5.1 is on the model atlas — Vals Index 67.87%

Anthropic's 1 Sep Mythos-class refresh of Fable 5. Same $10 / $50 list, cache reads cut to $0.25 per 1M tokens. Vals puts it at 67.87% on the Index, ahead of Opus 5 and Fable 5. AA Intelligence Index is not in the 1 Sep API snapshot yet. Compare it with GPT-5.6 Sol or Fable 5. /models/claude-fable-5-1

2026-09-01benchmarks

AA-Omniscience and CritPt join the benchmark atlas

Two missing slices of AA Intelligence Index v4.1 now have explainers and sourced rows: AA-Omniscience (factual recall vs hallucination — the knowledge primary) and CritPt (research-level physics, still under 40% at the frontier). /benchmarks/aa-omniscience · /benchmarks/critpt

2026-08-27benchmarks

Terminal-Bench-Science 0.1 is on the atlas — best row 30%

Stanford and Laude's 70-task science-agent suite: real lab workflows, still far from ceiling. Claude Opus 5 + Claude Code leads at 30.0%; GPT-5.6 Sol + Codex is 22.4%; GLM-5.3 is the strongest open-weight row at 8.1%. Not Terminal-Bench 2.1. /benchmarks/terminal-bench-science

2026-08-26models

GLM-5.3-Flash is on the model atlas — AA index 57.5

Z.ai's Flash sibling of GLM-5.3, from the 1 Sep Artificial Analysis snapshot: Intelligence Index 57.5 at $0.15 / $0.50 per 1M tokens. Two points behind GLM-5.3 max at a tenth of the list price. /models/glm-5-3-flash · /compare/models/glm-5-3-vs-glm-5-3-flash

2026-08-27guide

Three academic guide trios, rebuilt

Student researcher, literature reviewer, and peer reviewer — each with a matching stack and playbook. Peer review starts with the venue AI policy. Retired grant, coding, and deck paths redirect to /guides.

2026-08-24models

GLM-5.3 is on the model atlas — AA index 59.5

Z.ai's August coding refresh is on the atlas from the 2026-08-24 Artificial Analysis snapshot: max-effort Intelligence Index 59.5, API list $1.40 / $4.40 per 1M tokens. Same base as GLM-5.2; weights were still unpublished on this check. Ledger rows sync from AA; editorial depth still needs a desk pass. Compare it with Qwen3.8 27B. /models/glm-5-3 · /compare/models/glm-5-3-vs-qwen3-8-27b

2026-08-24models

Qwen3.8 27B is on the model atlas — AA index 52

Alibaba's dense open-weight Qwen3.8 sibling is on the atlas from the 2026-08-24 Artificial Analysis snapshot: xhigh Intelligence Index 52, hosted list $0.50 / $3 per 1M tokens, Apache 2.0 on Hugging Face. This is not hosted Qwen3.8 Max. Ledger rows sync from AA; editorial depth still needs a desk pass. Compare it with Qwen3.8 Max. /models/qwen3-8-27b · /compare/models/qwen3-8-27b-vs-qwen3-8-max

2026-08-15models

Qwen3.8 2.4T is on the model atlas — AA index 57.7

Alibaba's open-weight Qwen-Max-class MoE (2.4T total / 95B active) is on the atlas from the 2026-08-15 Artificial Analysis snapshot: Intelligence Index 57.7, hosted list $2 / $6 per 1M tokens. This is the published checkpoint, not hosted Qwen3.8 Max. Ledger rows sync from AA; editorial depth still needs a desk pass. Compare it with Qwen3.8 Max. /models/qwen3-8-2-4t-a95b · /compare/models/qwen3-8-2-4t-a95b-vs-qwen3-8-max

2026-08-13models

Gemini 3.7 Flash is on the model atlas — AA index 56

Google's August Flash workhorse is on the atlas from the Artificial Analysis snapshot: high-effort Intelligence Index 56, intro price $0.75 / $3.75 per 1M tokens through 31 Dec 2026. Ledger rows sync from AA; editorial depth still needs a desk pass. Compare it with 3.6 Flash or Claude Sonnet 4.6. /models/gemini-3-7-flash · /compare/models/gemini-3-6-flash-vs-gemini-3-7-flash

2026-08-12models

Grok 4.6 is on the model atlas — AA index 60.9

xAI's August flagship is on the atlas from the Artificial Analysis snapshot: high-effort Intelligence Index 60.9, $2 / $6 per 1M tokens. Ledger rows sync from AA; editorial depth still needs a desk pass. Compare it with GPT-5.6 Sol. /models/grok-4-6 · /compare/models/gpt-5-6-sol-vs-grok-4-6

2026-08-03desk

VerdictPal is live, and the Atlas Report puts numbers on the whole atlas

The launch note is published, plus a new original-research surface: the Atlas Report computes aggregate pricing, privacy, benchmark-saturation and freshness figures from every published tool, and it is free to cite with a link. Highlights this edition — most dossiers do not record any stance on training with your content, paid entry plans cluster around the median, and frontier model input prices spread by more than three orders of magnitude. /report · /launch

2026-07-12bench

AI Builder Design Test on the Lab bench — four builders, same prompts, very different results

VerdictPal Lab's second first-party protocol: four identical multi-turn design prompts through Lovable, v0, Replit, and Bolt. Run 01 recorded a Turn 1 ranking (Lovable → v0 → Replit → Bolt) and the cost-of-arrival in each vendor's native unit — no padded leaderboard, formal scored rows pending a repeatable rubric. /benchmarks/ai-builder-design-test · /benchmarks/lab

2026-07-06bench

Citation Fidelity v0.1 protocol frozen on the Lab bench

VerdictPal Lab's first first-party benchmark: twelve research questions, dual-reviewer citation grading, frozen runbook on the atlas page — protocol published, desk run still pending, no padded leaderboard. /benchmarks/citation-fidelity · /benchmarks/lab

2026-07-02guide

Five research guides now on the atlas

Citation audit, systematic review protocol, data cleaning pipeline, grant writer path, and peer reviewer path — solid guides with step blocks and verification checklists.

2026-06-25guide

Student research workflow on the guides pillar

Path, stack, playbook, and showcase for turning broad prompts into source-backed briefs — the workflow we point new readers at first.

2026-06-24tool

Factory flagship dossier shipped

Agent-native delivery with Droid, Spec Mode, missions, and enterprise controls — fit 74 with the autonomy and pricing caveats on the record.

2026-06-23tool

Cursor flagship dossier shipped

VS Code-compatible AI editor at fit 81 — Tab, Chat, Agent, Privacy Mode, team governance, and the diff-review failure modes we publish before you merge.

2026-06-22tool

Gemini flagship dossier shipped

Google's multimodal assistant ecosystem at fit 75 — Search grounding, Deep Research, storage bundles, and where it is not a cited search engine.

2026-06-21tool

NotebookLM flagship dossier shipped

Source-grounded reading workspace at fit 78 — corpus chat, Audio Overviews, Flashcards, and the empty-corpus trap documented on the dossier.

2026-06-20tool

Perplexity flagship dossier shipped

Cited search front door at fit 72 — consumer Pro/Max, Deep Research, Sonar APIs, and failure modes for bibliographies you have not opened.

2026-06-15tool

OpenCode flagship dossier shipped

The open-source coding agent at fit 95 — Plan/Build modes, Go and Zen pricing, 75+ providers, and the free-model privacy caveats on the record.

2026-06-15tool

Command Code flagship dossier shipped

Terminal coding agent with taste-1 learning, open-model deals from $1/mo, and the full pricing and privacy picture — including what lives in .commandcode/.

2026-06-15tool

Raycast flagship dossier shipped

The macOS launcher we install before debating another AI tab — extensions, Quick AI, clipboard, and honest failure modes at fit 97.

2026-06-08guide

Grant deadline set (archived)

The grant-deadline path has been retired. Academic guides now live as three trios: student researcher, literature reviewer, and peer reviewer.

2026-06-02guide

Trusted-source brief workflow shipped

The Guides pillar now has a real student research path, tool stack, playbook, and showcase for turning broad prompts into source-backed briefs.

2026-06-02design

Dark mode rebuilt for the evidence instrument

Dark surfaces now use muted amber, readable borders, and component-specific overrides instead of neon panels and inverted cream panels.

The above are real drop entries; there's no fixed schedule. When the atlas changes, you'll know.

In the next drop · preview

Three tools we're watching.

Subscribers see these promoted, retired, or updated first. On desktop, tool tiles tilt when you point at them.

Build & code

Command Code

88/100

ROLETaste-aware terminal coding agent and open-model harness

Terminal coding agent with continuous taste learning (taste-1), open-model deals, MCP/skills, and subscription credits from $1/mo — ships, tests, and refactors in your conventions.

Watch outYou need a GUI IDE with Tab completion as the primary surface — use Cursor instead

Editorial · hands-onFlagship

Developers who wantBuilders experimenting DeepSeek,

ROLEAI-native code editor and agentic refactor workspace

VS Code-compatible AI editor with Tab completion, Chat, Agent, MCPs, skills, hooks, cloud agents, Bugbot, Privacy Mode, and team/enterprise governance controls.

Watch outYou cannot enable Privacy Mode or get approval before sending sensitive code to AI providers

Editorial · hands-onFlagship

Developers already livingRepo-wide refactors where

ROLEAgent-native delivery surface

Agent-native software development platform around Droid, with terminal, desktop, CLI, SDK, local/cloud background agents, missions, review workflows, and enterprise controls for delegated software delivery.

Watch outYou only want lightweight VS Code autocomplete

Editorial · hands-onFlagship

Multi-step refactors, migrations,Teams wiring Droid

The Pack · Editorial newsletter

New tools in your inbox. Not a leaderboard.

When a tool ships, gets retired, or moves status, subscribers hear about it first. Affiliate links are disclosed when present, no third-party trackers in the email. One click to unsubscribe.

Sealed Pack is not open yet. Join The Pack free. Paid drops ship when the deliverables are concrete.

✓ One click to unsubscribe
✓ No third-party analytics in the email
✓ We never sell or rent the list

★

The Pack · free

The editorial newsletter. When a tool ships, gets retired, or changes status, you get one short, defensible note. Read in under four minutes.

◆

Sealed Pack · coming soon

Planned: everything in The Pack, plus early draft dossiers and short editor context when a tool changes. Signup records your interest, billing is not live yet. See the tier note in the form below.

Paid tier details stay unpublished until deliverables are concrete; we do not ship perks we cannot fulfill.

Why pay for any of this? Because VerdictPal does not sell verdicts. Affiliate links are disclosed when present and never affect scores; reader support keeps the desk independent. If you don't want to pay, the atlas stays free. If you can, that's how it stays honest.

The PackWhen a tool ships.

September 2026 frontier: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, Hy4 preview

September 2026 frontier: GPT-6 Astra, Gemini 3.8 Flash, Muse Spark 1.3, Hy4 preview

Claude Fable 5.1 is on the model atlas — Vals Index 67.87%

AA-Omniscience and CritPt join the benchmark atlas

Terminal-Bench-Science 0.1 is on the atlas — best row 30%

GLM-5.3-Flash is on the model atlas — AA index 57.5

Three academic guide trios, rebuilt

GLM-5.3 is on the model atlas — AA index 59.5

Qwen3.8 27B is on the model atlas — AA index 52

Qwen3.8 2.4T is on the model atlas — AA index 57.7

Gemini 3.7 Flash is on the model atlas — AA index 56

Grok 4.6 is on the model atlas — AA index 60.9

VerdictPal is live, and the Atlas Report puts numbers on the whole atlas

AI Builder Design Test on the Lab bench — four builders, same prompts, very different results

Citation Fidelity v0.1 protocol frozen on the Lab bench

Five research guides now on the atlas

Student research workflow on the guides pillar

Factory flagship dossier shipped

Cursor flagship dossier shipped

Gemini flagship dossier shipped

NotebookLM flagship dossier shipped

Perplexity flagship dossier shipped

OpenCode flagship dossier shipped

Command Code flagship dossier shipped

Raycast flagship dossier shipped

Grant deadline set (archived)

Trusted-source brief workflow shipped

Dark mode rebuilt for the evidence instrument

The Pack · free

Sealed Pack · coming soon

The Pack
When a tool ships.