Token Use: baixar versions anteriors

tokenuse v1.2.7 - Small Change
tokenuse 1.2.7, named Small Change, makes money read the same everywhere, down to the last penny. It adds names and prices for Claude Haiku 5.5 and Sonnet 5.5, repairs costs already archived for them, and fixes a few UI details on the TUI and desktop app. It also patches new dependency advisories, and Windows releases are now code-signed.
Money and pricing
- Money is formatted the same way in the TUI and desktop app: two decimals with thousands separators (£1,088.42, not £1088), £0 for exactly zero, and <£0.01 for amounts below a penny instead of £0.0000 or £0.0070. Desktop call rows and the call detail show the exact cost on hover, and the TUI call detail shows it inline. Zero-decimal currencies such as JPY show whole units.
- Claude Haiku 5.5 and Sonnet 5.5 now have their own model names and prices. Haiku 5.5 no longer shows as "Haiku 5 5" with a fallback-pricing warning, and Sonnet 5.5 is no longer shown as Sonnet 5 or charged Sonnet 5's higher cache-read rate. Haiku 5.5 prompts over 100,000 tokens are still estimated at the standard rate.
- Cursor's Gemini 3.7/3.8 Flash output rate is now $3.75/MTok, and the pricing books were refreshed on October 18, 2026.
Archive repair
- Archive schema v11 reprices Claude Haiku 5.5 and Sonnet 5.5 calls already in the archive. Haiku 5.5 calls were charged at the Sonnet 4.6 fallback rate, about 30 times too much. Claude Code Sonnet 5.5 calls were charged Sonnet 5's cache-read rate, which is twice the real one. A row changes only when its stored cost proves the old price was used, so correct rows are left alone and the repair never runs twice. Cursor and Copilot Sonnet 5.5 calls are not repriced, because those tools still price Sonnet 5.5 at the Sonnet 5 rate.
Interface fixes
- Coach colours its practice score bars and sparklines using the same thresholds as the grade letters (green from 90, yellow from 70), so an 80 no longer shows a green bar next to a yellow B. The TUI grade colours now use the same three tones.
- In ranked tables, the bar showing each row's share is now a lighter background with a thin underline. The old full-height edge looked like a text cursor. The TUI drops its brighter end cell the same way.
- The desktop page title updates when you move from one tool page to another. Previously the Codex page could still be titled "Claude Code".
- The desktop status bar only shows period, tool, sort, and project shortcuts on pages that have those controls, and the shortcuts only work there. The tool shortcut now also works on Coach. The TUI Coach footer no longer lists g sort.
Security and distribution
- Windows releases are now Authenticode-signed with a Certum code-signing certificate: the TUI's tokenuse-windows-amd64.exe, the desktop NSIS and MSI installers, and the app executable inside them. SmartScreen warnings should fade as signed releases build reputation.
- Updated rustls to 0.23.46 in both lockfiles (RUSTSEC-2026-0346 to 0349), rust_xlsxwriter to 0.100.0 with zip 9 (RUSTSEC-2026-0367 to 0369), and the desktop's transitive devalue and source-map-js packages.
- The desktop updater plugin is now built without its optional zip feature. Releases publish raw NSIS and AppImage updater files, so the old zip and tar.gz extraction code was never used, and it was the only thing pulling in the vulnerable zip 4.
Upgrade notes
- The first launch runs the archive v11 migration once. Haiku 5.5 history drops sharply to its correct price, and Claude Code Sonnet 5.5 history drops by the overcharged cache-read amount.
- Pricing books downloaded before this release are older than the embedded books, so the embedded books are used automatically.
- At release preparation, cargo-audit reported no known vulnerabilities in either lockfile, and pnpm audit reported none either. RustSec still reports upstream maintenance warnings, the Linux GTK dependency's glib unsoundness warning, and a yanked yoke-derive.

tokenuse v1.2.6 - Price Cut
tokenuse 1.2.6, named Price Cut, adds pricing for the models Anthropic and OpenAI released on September 22, 2026, both cheaper than the models they replace. It also corrects costs already archived for those models and refreshes the pricing books and dependencies.
New models
- Claude Opus 5.5 is priced at $4 input, $5 cache write, $0.20 cache read, and $20 output per MTok. Cache reads are 0.05x input, and Claude Code fast mode is 2x. The price applies to Claude Code, Cursor, and Copilot usage.
- Opus 5.5 now appears as its own model. Previously it was shown and grouped as Opus 5.
- GPT-6 Sol ($2/$10 per MTok) and GPT-6 Luna ($0.10/$0.50 per MTok) are priced for Codex and Copilot, using the default tier with 10% cache reads and 1.25x cache writes.
Pricing fixes
- Archive schema v10 corrects costs that were already stored for the new models. Opus 5.5 calls were charged at the inherited Opus 5 rates, and GPT-6 Sol, Luna, and Astra calls were charged at the Sonnet fallback rates. A row is recomputed only when its stored cost exactly matches the stale formula. Already-correct rows are left unchanged, and the hidden 1-hour cache-write premium is preserved.
- The v10 pass also corrects GPT-6 Astra rows in archives that reached schema v9 before the Astra repair was added to it.
- gpt-5-codex, gpt-5.1-codex (including Max and Mini), gpt-5.2-codex, and codex-mini-latest now use OpenAI's published rates. LiteLLM dropped their direct entries. Without this change, the regional Azure entries would have added 10%, and codex-mini-latest would have used fallback pricing.
- Pricing books were refreshed on September 23, 2026.
- Pricing refreshes now keep expired historical pinned rates without requiring their old marker text to remain on a provider's live page. Current pins still fail loudly when an entire configured source disappears.
Dependency updates
- Updated both Rust lockfiles with semver-compatible releases, including Tauri 2.11.6, the Tauri updater plugin 2.12.0, rustls 0.23.45, and ureq 3.4.2. This removes RUSTSEC-2026-0285 from the root lockfile.
- Updated Svelte, Vite, three.js, Motion, Lucide icons, and the Tauri CLI. TypeScript stays on 6.x. Moving to TypeScript 7 is a major upgrade and is left for a separate change.
Upgrade notes
- The first launch runs the archive v10 migration once. Totals for Opus 5.5, GPT-6 Sol, GPT-6 Luna, and affected GPT-6 Astra history go down to the correct prices.
- Pricing books downloaded before this release are older than the embedded books, so the embedded books are used automatically.
- Neither lockfile has known vulnerabilities reported by cargo-audit, and pnpm audit reported none at release preparation. RustSec still reports upstream maintenance warnings and the Linux GTK dependency's glib unsoundness warning.

tokenuse 1.2.5, named Fresh Start, updates the Rust and desktop dependencies, removes known vulnerable dependency versions, and adds security audits as release gates. It also fixes the WinGet release workflow's fork synchronisation.
Dependency updates
- Updated Tauri to 2.11.5 and refreshed its build tooling and plugins, including the fixes for the Windows/Android origin-confusion advisory in versions through 2.11.0.
- Updated the PDF renderer to Fulgur 0.40 and its PDF dependencies, addressing the advisories found in the previous locked versions. HTML/PDF exports now use its current rendering API.
- Applied the report's print stylesheet directly during PDF rendering and placed page breaks between sections, preventing a blank fourth page. The export test now checks the PDF's page count.
- Updated the remaining Rust dependencies and regenerated both lockfiles, including SQLite, Excel exports, terminal rendering, HTTP downloads, and OS credential storage.
- Migrated HTTP requests to ureq 3 while retaining the quota-sync authentication headers, optional endpoint handling, and 20-second request timeout.
- Updated Svelte, Vite, the Tauri frontend packages, fonts, icons, and supporting libraries. Replaced the deprecated lucide-svelte package with @lucide/svelte. TypeScript stays on the latest 6.x release supported by the current Svelte checker.
Release safeguards
- CI now audits both Rust lockfiles against RustSec and audits the desktop lockfile with pnpm, including development dependencies. Failed audits block release builds.
- AGENTS.md requires the full root and desktop checks and all three security audits before each release, followed by verification of the published assets and distribution jobs.
- Tagged releases publish their checked-in title and release notes automatically and require a matching release notes file.
- Fixed WinGet fork synchronisation by using gh repo sync and checking that the configured classic PAT has both public_repo and workflow scopes.
Upgrade notes
- No new archive migration or configuration format change is introduced by this release.
- The updated lockfiles have no known vulnerabilities reported by cargo-audit and no advisories reported by pnpm audit at release preparation. RustSec still reports upstream maintenance warnings and the Linux GTK dependency's glib unsoundness warning; these remain visible in CI for review.
- Windows executables and installers remain unsigned with Authenticode. Tauri updater signatures protect updater downloads but do not establish Windows publisher reputation. Dependency updates alone do not establish the cause of the reported antivirus detection or guarantee it is resolved.

tokenuse` 1.2.4, named Price Check, refreshes the rate card across OpenAI, Anthropic, Cursor, and GitHub Copilot after a busy run of model launches and pricing changes. It adds exact identities and prices for the newest models, keeps dated price schedules from rewriting older usage, and strengthens the maintainer refresh so regional uplifts, non-chat placeholders, and incomplete rows cannot quietly become the default. A guarded archive migration corrects the small set of historical calls that can be proven to have used a stale or fallback rate.
This release also adds WinGet as the supported package path for the Windows desktop app. New releases continue to ship the usual GitHub assets and now automatically open the matching microsoft/winget-pkgs manifest pull request for the MSI.
Highlights
- Current model coverage: exact rows and friendly identities now cover GPT-6 Astra, Claude Fable 5.1 and Mythos 5.1, Grok 4.5/4.6, Gemini 3.6/3.7/3.8 Flash, MAI-Code-1.1-Flash, Kimi K2.7 Code/K3, and GLM 5.2 across the tools that expose them.
- Time-aware rates: GPT-5.6 Sol switches to its August 21 $4/$20 promotion and 1.25x cache-write charge without repricing earlier calls; unsuffixed gpt-5.6 usage resolves to Sol. Luna falls to $0.20/$1.20 and Terra to $2/$12 per MTok. GPT-6 Astra uses $10 input, $1 cached input, $12.50 cache write, and $50 output per MTok.
- Claude corrections: Fable/Mythos 5.1 cache reads use $0.25/MTok instead of the $1 predecessor rate, while Sonnet 5 keeps its permanent $2 input/$10 output launch price rather than the canceled September increase.
- Safer local upgrades: a newer embedded pricing book wins over an older downloaded pair, so stale Config-page files no longer force new models such as GPT-6 Astra through fallback pricing.
- Guarded history repair: archive schema v9 fixes only rows whose stored cost exactly proves one of the known bad schedules; ambiguous, already-correct, and unrelated calls stay untouched.
- Windows package management: winget install --id RussMckendrick.TokenUse -e installs the desktop MSI, and the release workflow submits each new version to WinGet automatically.
Provider and model updates
OpenAI's official model Markdown is now a first-class refresh source. GPT-6 Astra receives an exact row instead of inheriting the fallback, GPT-5.6 Sol carries its effective-dated promotional schedule, GPT-5.6 Luna drops to $0.20 input and $1.20 output per MTok, and GPT-5.6 Terra drops to $2 input and $12 output. GPT-5.4's existing $2.50 input, $0.25 cached input, and $15 output rates are now pinned to its official model page rather than changed. Dated model snapshots remain separate whenever their rate differs, so a newer family price never leaks backward into archived usage.
Anthropic's live pricing table supplies current Claude rows and corrects the historical Claude 3.5 Haiku schedule for newly imported or rebuilt history. Fable 5.1 and Mythos 5.1 are named independently from their 5.0 predecessors and use their published cache-read discount. Sonnet 5's canceled September 1 increase is not represented as a live schedule.
Cursor's first-party table is scraped again rather than carried only as a static emergency pin. Grok 4.6, Cursor-specific Gemini 3.8, GLM 5.2, Kimi K2.7 Code, and GPT-5 Fast pricing are included, while removed Composer prices remain available as historical rows. Explicit Fast rows cover GPT-5.4 at 2x, GPT-5.5 at its published $12.50 input/$75 output rate, and GPT-5.6 Luna/Sol/Terra at 2x including cache writes; the fixed web-search fee stays $0.01. Structured Cursor ids with a numeric context above 272K select the matching GPT-5.4/5.6 long-context rate, while exactly 272K stays on the default tier. The legacy $1.25/$6 Enterprise Auto schedule expires on September 7, 2026; newer unresolved default calls are marked fallback-priced because Cursor bills the routed model and the local artifact does not reliably identify it. Plan-specific token, regional, and legacy Max uplifts remain 

tokenuse 1.2.3, named Back Pay, teaches the pricing books about Claude Opus 5 and then settles the bill for the calls that were already made. Opus 5 matched no pricing row until now, so every Opus 5 call quietly fell through to the Claude Sonnet 4.6 fallback and was billed at $3/$15 per MTok instead of $5/$25 — an under-count of roughly 40% that no warning surfaced, because an unpriced model is indistinguishable from a cheap one once the fallback has done its work. This release prices the model properly, names it across every surface, and includes a one-shot archive repair that puts the missing cost back into history.
Highlights
- Claude Opus 5 is priced and named: $5/MTok input and $25/MTok output, with 5-minute cache writes at $6.25, cache reads at $0.50, and fast mode at 2x base. It appears as "Opus 5" in the Opus family across the dashboard, reports, and model pages.
- History is repaired automatically: archive schema v8 corrects Opus 5 costs already frozen into archive.db, once, the first time this version starts. Existing Opus 5 totals will rise; nothing else is touched.
- Copilot coverage too: GitHub's Opus 5 rate is tracked under the copilot tool scope, alongside the existing per-provider Copilot rows.
- A broken pricing refresh is fixed: Cursor restructured its documentation and removed the table the refresh keyed on, which was failing the entire refresh run rather than just that one source.
Opus 5 pricing
The curated Anthropic rows now carry Claude Opus 5 with a 2x fast_multiplier, matching Claude Code fast mode's published $10/$50 per MTok. The LiteLLM import needed no change: the existing claude-opus- filter picked up claude-opus-5 and its Bedrock and Google Cloud regional variants on the next refresh. The shared model registry gains an Opus 5 rule ahead of the claude-opus-4* prefixes, so the point-release-free id is not shadowed by its predecessors, and dated or vendor-prefixed ids fold into it as usual.
Claude Opus 4.7 fast mode was removed upstream on July 24, 2026. Its 6x fast_multiplier row is deliberately kept so archived fast-mode calls from before the removal still price correctly.
Repairing archived costs
Archived costs are frozen at import time on purpose — the archive inherits import-time cost precisely so that history is never silently repriced — so this correction is a deliberate, versioned migration rather than an ongoing reprice pass. It runs once, gated on the schema version, and cannot re-run.
The repair rescales affected rows rather than recomputing them. Opus 5's input, output, cache-write, and cache-read rates are each exactly 5/3 of the Sonnet 4.6 rates the calls were charged at, so a single factor restores the true cost. That approach is not just simpler than a recompute, it is more accurate: the 1-hour cache-write premium scales with the cache-write rate but is never persisted on the row, so a recompute would silently drop it. Web-search requests bill at $0.01 under both rows and are held out of the rescale, and speed = 'fast' rows additionally pick up the 2x multiplier the fallback row never carried.
Every row is checked before it is touched. A row whose stored cost already reflects Opus 5 pricing is skipped, so anyone who downloaded corrected pricing books before upgrading the binary cannot be double-charged. Model ids are matched through the shared canonicalization, so dated and vendor-prefixed ids are included while a distinct neighbour such as claude-opus-5-fast is not.
Pricing refresh maintenance
Cursor folded its standalone "Auto pricing" table into the combined "Model pricing" table on the same documentation page. Because a markdown source whose heading no longer exists fails the whole refresh rather than only its own rows, this had taken --refresh-prices out of service entirely. Cursor Auto is now read from that table's Auto Cost row; its rates are unchanged at $1.25 input and cache write, $0.25 cache read, and $6 output.
Upgrade notes
- The archive schema moves from v7 to v8. The mi