## (daily refresh)

- No entrant cleared the bar (okf-agent-memory re-reviewed and rejected again); no new note added [glm-5.3-flash]
- Harnesses: amp's September news folded in; claude-code open issues+PRs down to about 12.9k with two feature-timeline months corrected; deepseek-harness 215,497 stars with dsh-v0.1.3-alpha.2; fx v0.0.8 with breaking shell and subagent changes; onecli v2.6.0; zerostack v1.8.4; aider's stall re-confirmed [glm-5.3-flash]
- Exo: corrected the FrontierHarness sentence to nine harnesses; an earlier rewrite to twelve was wrong [glm-5.3-flash]
- Surfaces and orchestration: omnara now publishes usage-based pricing ($0 platform fee, $0.0414 per GiB-hour machine runtime) and joined as an eighth source; paseo 16,440 stars with open issues+PRs dropping 1,241 to 816; vscode-copilot and windsurf references canonicalized [glm-5.3-flash]
- Skills, context, and executions: agentsview's 84-223x ccusage benchmark removed from its docs (prose, caution, and reference revised); graphify v0.9.56 at 115,802 stars; anthropic-structured-outputs docs now list Opus 4.7/4.8 and Mythos Preview [glm-5.3-flash]
- Memory, control, and review: paperclip 80,209 stars and 5,392 issues; graphite.dev now redirects to graphite.com (five vendor reference URLs and matrix wording updated); open-code-review v1.11.6; deepeval docs moved to deepeval.com [glm-5.3-flash]
- Sandbox, factory, spec, and runtimes: har v1.14.1; ouroboros v0.54.1; tessl published platform pricing for the first time (Free $0 with 1,000 credits, Team $100 per month, Enterprise custom) [glm-5.3-flash]
- Essays and matrices: model-selection-for-coding-tasks re-fetched all sixteen tabled prices with zero moves (the GLM-5-Turbo clause dropped, it no longer appears on Z.ai's pricing page); the landscape as-of bumped; all 20 matrices re-verified and matched to their membership [glm-5.3-flash]

## (owner-prompted: model-selection table transposed)

- Owner request in chat: the model-selection-for-coding-tasks lineup table transposed, models moved from sixteen rows to sixteen columns so it reads as two attribute rows (input/output per 1M, notes) against the model columns, prices and column order unchanged [glm-5.3-flash]

## (owner-prompted: table property rows)

- Owner request in chat: more properties in the transposed lineup table; the Notes column decomposed into four property rows (Cached input per 1M, Context behavior, Pricing windows, Notes), grounded in the article's verified prose with no new facts, and a ? cell plus footnote marks properties the pricing pages do not state [glm-5.3-flash]

## (owner-prompted: one-year release window)

- Owner request in chat: keep only models released within the last year; all sixteen lineup entries pass the 2025-09-08 cutoff, so no column was dropped, and a Released row with month-precision cells plus a footnote states the one-year rule and the oldest entry [glm-5.3-flash]

## (owner-prompted: GLM-5.3-Flash added)

- Owner request in chat: include GLM-5.3-Flash; an eighteenth column added between GLM-5.3 and deepseek-v4-pro (released 2026-08-26, $0.075/$0.25 under a 50 percent promotion, cached reads $0.015, promo ends 2026-09-09), with the cheap-tier, inline-completion, price-category, and high-volume decision bullets updated [glm-5.3-flash]

## (owner-prompted: GPT-6 Astra added)

- Owner request in chat: include GPT-6 Astra; a nineteenth column added after gpt-5.3-codex (released 2026-09-03, $10/$50, cached reads $1, context doubling past the long-context threshold), and the frontier decision bullet now leads with gpt-6-astra [glm-5.3-flash]

## (owner-prompted: Qwen models added)

- Owner request in chat: include Qwen models; two columns joined the challengers (qwen3.8-max and qwen3.8-flash, from Alibaba's Model Studio billing page and the OpenRouter models API, releases 2026-08-03 and 2026-08-26), cached-rate cells left ? per the unpublished discount rate [glm-5.3-flash]

## (owner-prompted: model provider matrix)

- Owner request in chat: a matrix comparing the model providers; created agents/model-provider-feature-matrix with seven provider columns sorted alphabetically and twelve rows [glm-5.3-flash]
- Column headers link to the canonical pricing pages and every cell traces to the model-selection guide's verified table or to pages fetched this run, since no provider notes exist; listed first under Comparison matrices, with model-selection-for-coding-tasks cross-linking it [glm-5.3-flash]

## (owner-prompted: models.dev as reference)

- Owner request in chat: use models.dev as the model list reference; all 20 tabled models appear there, which carries ids, release dates, and context windows but no prices, so the Released row was refined to exact models.dev dates and the four ? context cells were filled [glm-5.3-flash]
- Folded in gemini-3.8-flash as a sibling sharing the 3.7/3.6 intro pricing of $0.75/$3.75 through 2026-12-31; model-provider-feature-matrix: Zhipu's long-context cell moved to 1M flat, widening the flat-1M club to five vendors [glm-5.3-flash]