What theopenco/llmgateway shipped
Written by FoxPlug from the repository's own releases; not affiliated with Llmgateway. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/theopenco/llmgateway; Llmgateway did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of November 2, 2026
What shipped
- Added usage receipts so callers get the cost of every request across endpoints, binary speech replies, streams and video jobs. Pull request #4996
- GET /v1/models now exposes full pricing, including context-length tiers, peak rates, image token counts per resolution, regional pricing, service tier multipliers and active discounts. Pull request #4952
- Added public provider status pages, with a status page listing each provider's official status and uptime measured on real traffic. Pull request #4992
- Expanded Azure and realtime audio support, adding missing Azure mappings, honoring credential base URLs and streaming speech responses. Pull request #4963
- Image requests now accept complexity, GIF animation settings and edit markup, and detect GIF output bytes while preserving base64 and MIME. Pull request #4982
- Requests pinned to a deactivated provider mapping are now rejected instead of being silently rerouted. Pull request #4962
- Seedance video billing now uses completion tokens with exact resolution, input-video and generated-audio rates across all five BytePlus mappings. Pull request #4961
Why it matters
This week focused on making costs visible and correct: usage receipts, fuller pricing in /v1/models, and fixes to image, video and Google image token accounting mean callers can price and bill requests more accurately. Routing also became stricter, with pins to deactivated mappings rejected rather than silently rerouted, and public status pages make provider uptime visible.
Changelog entry
- Add usage receipts carrying request cost across endpoints, streams, binary replies and video jobs. Pull request #4996
- Expose full pricing in /v1/models, including tiers, peak rates, image token counts, regional pricing and discounts. Pull request #4952
- Add public provider status pages showing official status and measured uptime. Pull request #4992
- Forward image complexity, GIF settings and edit markup, and detect GIF output bytes. Pull request #4982
- Bill Seedance from completion tokens with exact resolution, input-video and generated-audio rates. Pull request #4961
- Reject requests pinned to deactivated provider mappings instead of rerouting them. Pull request #4962
- Expand Azure mappings and realtime support, including streamed speech responses. Pull request #4963
- Classify normalized stop reasons so successful responses are not flagged as unknown finish reasons. Pull request #4980
This week: usage receipts for every request, full pricing exposed in /v1/models, public provider status pages, SVG output from Image Studio, and stricter handling of pins to deactivated mappings.
Week of October 12, 2026
What shipped
- Better-auth was bumped to 1.7.10 to fix a critical magic-link account-takeover advisory. Pull request #4711
- Provider keys now require the Enterprise plan to set a custom base URL, closing a gap that let any plan point the gateway at an organization-controlled host. Pull request #4697
- The gateway now guards provider-issued download URLs, so tenant-supplied base URLs under BYOK can no longer be followed without checks. Pull request #4688
- Tool strict,
allowed_callersand result image fields are kept through parsing, andparallel_tool_callsis now honored on the responses path. Pull request #4669 - Requests that send both
allowed_domainsandblocked_domainstoweb_searchnow get a 400 instead of silently dropping one. Pull request #3896 - Requests using the legacy custom/ model prefix are now rejected with a clear error instead of taking a degenerate path. Pull request #3897
- Global admin stats can now break traffic down by model family instead of summing provider rows by hand. Pull request #4712
- The admin token load view splits counts into input, cache and output so cached input is visible. Pull request #4701
Changelog entry
- Bump better-auth to 1.7.10 to fix a critical magic-link account-takeover advisory. Pull request #4711
- Require the Enterprise plan to set a custom base URL on a provider key. Pull request #4697
- Guard provider-issued download URLs before the gateway fetches them. Pull request #4688
- Return 400 when
web_searchis given bothallowed_domainsandblocked_domains. Pull request #3896 - Reject the legacy custom/ model prefix with a clear error. Pull request #3897
- Keep tool strict and
allowed_callersthrough parsing and preserve result images. Pull request #4669 - Map the
tool_usestop reason in streaming without a duplicate stop chunk. Pull request #4668 - Add a model family view to global admin stats. Pull request #4712
- Split admin token load into input, cache and output. Pull request #4701
- Show compact token numbers by default in admin load views. Pull request #4702
Weekly update: conflicting domain filters and the legacy custom/ prefix now return clear errors, custom base URLs need Enterprise, and provider download URLs are guarded. Tool strict, allowed_callers and parallel tool calls survive parsing. Global stats break down by model family
Week of September 28, 2026
What shipped
- Daily sync worker now automatically enables models on managed credentials for Azure and AWS Bedrock when they are deployed in accounts, eliminating manual re-testing. Pull request #4401
- Soft rate limit mode added to preserve session pins to capped providers instead of silently re-pinning and losing warm prompt cache. Pull request #4402
- Audio and document requests now skip deactivated model mappings when searching the full catalogue to prevent 404 errors. Pull request #4430
- Provider credential insights moved from dialog to a dedicated page with error history, error split, and linkable statistics. Pull request #4414
- Incidents pages now toggle between grouping errors by model or by error type to surface non-model-specific failures. Pull request #4393
- Chat send on iOS app now instant with message and thinking indicator showing immediately; composer clears right away. Pull request #4427
- Streaming requests now fall back to another provider when a stream fails mid-response after returning 200. Pull request #4422
Why it matters
This week brings production reliability improvements for streaming, automatic credential updates for managed providers, and expanded model availability on major platforms. The fixes address long-standing issues with error handling, session pinning, and model routing that directly impact users of the gateway.
Changelog entry
- Feat(worker): Daily provider key model sync automatically enables deployed models on Azure and AWS Bedrock credentials Pull request #4401
- Feat(gateway): Soft rate limit mode preserves session pins to capped providers Pull request #4402
- Fix(gateway): Audio and document requests skip deactivated model mappings Pull request #4430
- Feat(admin): Provider credential insights moved to dedicated page with error history Pull request #4414
- Feat(airside): Incidents can toggle grouping by model or by error type Pull request #4393
- Fix(mobile): Chat send now instant on iOS app with immediate feedback Pull request #4427
- Fix(gateway): Streaming requests fall back on mid-response failures after 200 Pull request #4422
- Fix(models): DeepInfra Gemma 4 31B input price updated to $0.20 per Mtoken Release
- Fix(models): DeepInfra Hy3 deactivated effective 2026-10-08 Pull request #4385
Llmgateway ships daily credential syncing for managed providers, soft rate limits to preserve session pins, and expanded AWS Bedrock models. Streaming now falls back on mid-response failures.
Week of September 21, 2026
What shipped
- Configurable auto routing now uses a classifier to pick between models based on request content, replacing the previous hardcoded three-model cheapest strategy. Pull request #4213
- New Gateway Load page shows real-time request volume, latency, and error rates by organization, project, API key, model, or provider. Pull request #4241
- Routing savings insights let organizations see what smart or dynamic routing is worth to them by comparing actual costs against fixed baselines. Pull request #4239
- Verification run history now tracks every preflight test result, showing which carrier fixed failures and when. Pull request #4226
- Carriers can now temporarily pause live models without delisting them permanently. Pull request #4232
- Unit test suite now runs in parallel across files, cutting local execution time from 24 minutes to under 10 minutes. Pull request #4237
- Documentation split into separate products (DevPass, Lounge, Airside, LLM Gateway) with dedicated sidebars and OG images. Pull request #4230
- Product comparison pages refreshed against vendor sources checked 2026-09-27 for pricing, quotas, availability, and feature claims. Pull request #4262
Changelog entry
- Auto routing classifier: model selection now considers request content instead of hardcoded cost minimization Pull request #4213
- Gateway Load page: real-time request volume, latency, and error rates by organization, project, key, model, or provider Pull request #4241
- Routing savings insights: organizations can see what smart or dynamic routing is worth against fixed baselines Pull request #4239
- Error rate by model on Gateway Load: new error column on both canonical and mapping views Pull request #4248
- Latency insights on Gateway Load: latency rollups for organizations, projects, and API keys Pull request #4245
- Azure Priority service tier forwarding for low-latency Foundry processing Pull request #4221
- Verification run history: track all preflight test results with timestamps and trigger source Pull request #4226
- Pause models: carriers can temporarily stop traffic without permanent delisting Pull request #4232
- Bill smart routing classifier: Jev classification requests now appear in billing Pull request #4231
- Filter logs by user seat email in admin dashboard and organization-wide log view Pull request #4227
- Azure Priority processing changelog entry with billing behavior Pull request #3960
- Group unstable mapping errors by provider key to isolate credential-specific failures Pull request #4257
- Unit test suite now runs files in parallel; local execution ~10 minutes vs 24 minutes Pull request #4237
- Playwright tests sharded across 3 CI runners for faster feedback Pull request #4249
- Documentation split by product (DevPass, Lounge, Airside, LLM Gateway) with separate sidebars Pull request #4230
- Product comparisons refreshed against vendor sources (2026-09-27) for pricing, quotas, and features Pull request #4262
- Enterprise-first homepage redesign with clear next steps for company-wide rollout Pull request #4261
Load insights now show request volume, latency, and errors by org/project/key. Configurable auto routing picks models by request content. Routing savings reveal what smart routing is worth. Verification history tracks every preflight test.
Week of September 14, 2026
What shipped
- Native iOS app for Lounge shipped with full React Native implementation and backend support. Pull request #4082
- Organization owners can now delete organizations from settings with safeguards against in-use deletions and Stripe subscription cleanup. Pull request #4010
- Added typed-decision model support to gateway so Jev can be used for calibrated probability responses. Pull request #4156
- Playground image generation performance improved with lazy loading, zip downloads, and flex layout. Pull request #4163
- Routing analytics now break compliance exclusions and rate limits down by specific rule instead of single totals. Pull request #4121
- Admin model stats now display provider cache rates alongside cached token counts. Pull request #4136
- Custom signup email-domain blocklist moved to system settings table for admin management without code changes. Pull request #4141
- Global provider rate limits now accept zero RPM/RPD to block matching requests before upstream calls. Pull request #4140
Why it matters
This week brings major infrastructure work: a native iOS client for Lounge with full feature parity, organization deletion with proper safeguards, and broader model coverage including typed-decision models and improved parameter forwarding. Admin analytics and routing visibility also improved significantly across multiple surfaces.
Changelog entry
- Native iOS app for Lounge with React Native, device sign-in, profiles, leaderboards, Canvas, and Sandbox Escape. Pull request #4082
- Organization deletion from settings with safeguards and Stripe subscription cleanup. Pull request #4010
- Typed-decision model support for Jev through gateway. Pull request #4156
- Playground image preview lazy loading and zip download with flex layout. Pull request #4163
- GPT Image 2.5 added via Azure with two missing variant mappings. Pull request #4147
- Routing analytics compliance and rate limit exclusions broken down by specific rule. Pull request #4121
- Admin stats show provider cache rates alongside cached token counts. Pull request #4136
- Signup email-domain blocklist moved to system settings for admin management. Pull request #4141
- Global provider rate limits accept zero RPM/RPD to block requests before upstream. Pull request #4140
- Provider credential errors now show per-model split in tooltip. Pull request #4160
- 24h error rate per credential added to Provider Credentials table. Pull request #4145
- Airside carriers page links directly to carrier model mappings and traffic. Pull request #4122
- Admin cost charts support hourly/daily bucket override via query parameter. Pull request #4119