PromptFuel v1.0.24

Included in this release:

• Added GPT-6.1 Terra (gpt-6.1-terra) API-equivalent estimates effective 2026-10-26: $1.00 input, $5.00 output, $0.05 cached input, and $1.25 for 30-minute cache writes per million tokens. Estimates use standard short-context rates; 12-hour cache writes and long-context or service-tier modifiers are not represented.

PromptFuel v1.0.23

Included in this release:

• Added effective-dated API-equivalent pricing for Claude Haiku 5.5 (claude-haiku-5-5) at $0.10/1M input, $0.50/1M output, $0.125/1M 5-minute cache write, $0.20/1M 1-hour cache write, and $0.01/1M cache read, effective 2026-10-07. Before this change, Haiku 5.5 usage was not recognized and was estimated with the fallback Claude rate.
• Added an effective-dated Claude Sonnet 5.5 (claude-sonnet-5-5) row reflecting Anthropic's cache-read price reduction from $0.20/1M to $0.10/1M, effective 2026-10-07. All other Sonnet 5.5 rates are unchanged, and the earlier row is preserved so estimates dated before 2026-10-07 keep the $0.20/1M cache-read rate.
• Re-audited the Claude and OpenAI rate tables against the official provider pricing, model, deprecation, and changelog pages; no other rate changes were found since the previous refresh.

Scope notes:

• Claude Haiku 5.5 is priced in two bands: the rates above apply to requests whose prompt is up to 100,000 tokens, and requests with longer prompts (counting cache reads and writes) are billed at higher rates. PromptFuel estimates from aggregated per-model token totals and does not see per-request prompt length, so only the lower band is represented and Haiku 5.5 usage above the threshold is underestimated.
• Claude Mythos 5 and 5.1 are limited-availability and are not included, and no claude-opus-5-fast row is included because no verified effective date was found.
• Pricing remains API-equivalent estimates and does not model batch, flex, priority, ultrafast, long-context, or regional modifiers.

PromptFuel v1.0.22

Included in this release:

• Added effective-dated API-equivalent pricing for Claude Fable 5.1 (claude-fable-5-1) at $10.00/1M input, $50.00/1M output, $12.50/1M 5-minute cache write, $20.00/1M 1-hour cache write, and $0.25/1M cache read, effective 2026-09-01. Cache reads on Fable 5.1 are priced lower than on Claude Fable 5.
• Added effective-dated API-equivalent pricing for Claude Sonnet 5.5 (claude-sonnet-5-5) at $2.00/1M input, $10.00/1M output, $2.50/1M 5-minute cache write, $4.00/1M 1-hour cache write, and $0.20/1M cache read, effective 2026-09-28.
• Added effective-dated OpenAI API pricing for GPT-6.1 Sol (gpt-6.1-sol) at $2.00/1M input, $10.00/1M output, $2.50/1M cache write, and $0.10/1M cached input, effective 2026-09-29. GPT-6 Sol remains a separate entry at its existing rates. As with the other GPT-6 models, the flat pricing schema represents only short-context pricing; requests above 272K input tokens have a separate long-context tariff.
• Corrected Claude Sonnet 5 pricing: the scheduled 2026-09-01 increase to $3.00/1M input and $15.00/1M output was cancelled by Anthropic, so Sonnet 5 stays at $2.00/1M input and $10.00/1M output. Estimates dated on or after 2026-09-01 were previously overstated.
• Updated the model notes for retired or deprecated Claude models (Opus 4, Opus 4.1, Sonnet 4, and Sonnet 4.5) to match current Anthropic lifecycle status. Their pricing is unchanged.

PromptFuel v1.0.21

Included in this release:

• Added effective-dated API-equivalent pricing for Claude Opus 5.5 (claude-opus-5-5) at $4.00/1M input, $20.00/1M output, $5.00/1M 5-minute cache write, $8.00/1M 1-hour cache write, and $0.20/1M cache read, plus the matching Fast-mode label at $8.00/1M input and $40.00/1M output, effective 2026-09-22.
• Added effective-dated OpenAI API pricing for GPT-6 Sol (gpt-6-sol) at $2.00/1M input, $10.00/1M output, $2.50/1M cache write, and $0.20/1M cached input, and GPT-6 Luna (gpt-6-luna) at $0.10/1M input, $0.50/1M output, $0.125/1M cache write, and $0.01/1M cached input, effective 2026-09-22. As with GPT-6 Astra, the flat pricing schema represents only short-context pricing; requests above 272K input tokens have a separate long-context tariff.

PromptFuel v1.0.20

Included in this release:

• Large token and metric counts now display using B for billion-scale values (e.g. 1.5B) instead of continuing to display as thousands of millions (e.g. 1,500.0M), affecting both the TypeScript token count formatter and the webview metric number formatter.

PromptFuel v1.0.19

Included in this release:

• Added API-equivalent pricing estimates for GPT-6 Astra (gpt-6-astra) at $10.00/1M input, $50.00/1M output, $1.00/1M cached input, effective 2026-09-03, sourced from official OpenAI API pricing. Long-context pricing above 272K input tokens is not represented by the flat pricing schema.
• Added effective-dated GPT-5.6 Terra price reduction ($2.00/1M input, $12.00/1M output, $0.20/1M cached input) effective 2026-07-30, and GPT-5.6 Luna price reduction ($0.20/1M input, $1.20/1M output, $0.02/1M cached input) effective 2026-07-30, using the existing scheduled-pricing mechanism. Historical rows (effective 2026-07-09) are preserved so earlier dates continue to estimate correctly.

PromptFuel v1.0.18

Included in this release:

• Added two new settings governing which provider usage meters are displayed: promptFuel.showExtraUsage (default true) controls the Extra usage meter for any provider that reports one, and promptFuel.visibleUsageMeters (default [], an empty array) opts in individual generic provider meters by their normalized ID, or accepts "*" to show every generic meter currently reported. "*" covers generic meters only and does not re-enable Extra usage when promptFuel.showExtraUsage is false. Primary quota windows (5h/7d) are unaffected by either setting.
• Added the PromptFuel: Show Available Usage Meters command, which lists every currently reported meter ID (including hidden ones) from already-loaded state, without triggering a provider request, and copies a selected ID to the clipboard for pasting into promptFuel.visibleUsageMeters.
• Changed generic provider usage meters — any meter that is not a primary 5h/7d window and not Extra usage — to be hidden by default and shown only when their normalized ID is listed in promptFuel.visibleUsageMeters. Extra usage keeps its existing value-based behavior of hiding at exactly 0% and when fully exhausted, unaffected by this change.
• Changed Claude's seven_day_opus field to no longer receive dedicated live handling; it now flows through the same generic meter parsing path as any other provider-reported field, which means it is hidden by default like any other generic meter. To keep showing it, set "promptFuel.visibleUsageMeters": ["seven-day-opus"]. Cached quota data and machine snapshots written by earlier versions still read correctly, but carry the older ID claude-seven-day-opus until a live refresh replaces them; list both IDs to cover cached, snapshot, and mixed-version cases.

PromptFuel v1.0.17

Included in this release:

• Added effective-dated GPT-5.6 Sol promotional pricing ($4.00/1M input, $0.40/1M cached input, $20.00/1M output) effective 2026-08-21, using the existing scheduled-pricing mechanism. The pre-promotion row ($5.00/1M input, $0.50/1M cached input, $30.00/1M output, effective 2026-07-09) is preserved as historical pricing. The promotion is stated to remain available at least through 2026-11-21; no speculative reversion row was added. Pricing remains short-context only — requests above 272K input tokens have a separate long-context tariff not represented by this flat pricing schema.

PromptFuel v1.0.16

Included in this release:

• Improved reset countdown formatting to omit empty time units and show the next meaningful unit instead — for example, 5d27m instead of 5d0h; exact-day and exact-hour values stay compact, and positive durations under one minute display as <1m.
• Hid Claude's Extra usage meter from the status bar and tooltips when it is fully exhausted, while preserving the existing behavior that also hides the meter when Extra usage has not been used yet. Extra usage remains visible while it has been partially consumed and still has quota remaining; other meters, including Codex, are unaffected.

PromptFuel v1.0.15

Included in this release:

• Made startup snappier by publishing quota and reset status before the slower local history discovery and scanning work, so the status bar reflects usage right away; the Codex foreground path no longer waits for the full local session scan before showing its initial status.
• Added progressive history loading in stages — 1 day, 7 days, 30 days, 365 days, then all available history — with the dashboard noting when only part of the history window has loaded so far. The ALL chart still shows the most recent 12 months of the loaded history window to keep it readable.
• Persisted per-file history contributions across VS Code sessions so unchanged history files are reused instead of reparsed on every load; changed and deleted files are still detected and rescanned.
• Made warm refreshes efficient: when valid full history is already cached, a refresh re-checks the whole window in a single pass instead of replaying the full progressive sequence, reducing unnecessary dashboard and status work and avoiding briefly showing a narrow or partial history window during an ordinary refresh.
• Skipped building expensive dashboard models while the Usage dashboard is closed, kept the history-loading progress banner consistent and anchored to the bottom of the History section, and added a one-time notification when a genuine first history load finishes — not on every VS Code startup when a valid persisted cache already exists.
• Made reset countdowns more precise: minutes under an hour, hours and minutes under a day, and days and hours at a day or more (for example 1h55m or 2d5h), while partial remaining minutes are still rounded upward.
• Hid Claude's disabled or unused Extra usage meter from the status bar and tooltips when it is exactly 0%, while nonzero Extra usage remains visible; other meters, including Codex, are unaffected.
• Stopped treating live Codex tracing and context-window token counters as dated correlated usage history. Codex history buckets without turn or model evidence are now rejected when creating snapshots, reading self archives, importing remote/shared history, and building combined dashboard totals, preventing false large token totals with zero turns and no model attribution, including totals combined across shared machines.
• Ignored already-written malformed snapshot and archive data: the corrected behavior leaves it out of display and stops it from being perpetuated, and existing snapshot and archive files are not automatically rewritten or deleted. No manual data cleanup is required for normal corrected behavior once all participating machines are updated.