What opencost/opencost shipped
Written by FoxPlug from the repository's own releases; not affiliated with Opencost. 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/opencost/opencost; Opencost 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 October 26, 2026
What shipped
- Fixed the output processing time calculation so decode duration is no longer read from a metric that does not measure it. Pull request #4154
- Fixed GPU request, usage, and shared-GPU detection for clusters and metrics sources that report raw Kubernetes resource names like nvidia.com/gpu instead of the Prometheus-sanitized form. Pull request #4134
- Release images and SBOMs are now built from the exact vX.Y.Z tag for both tag pushes and manual runs, instead of scanning develop HEAD after release. Pull request #4100
- Missing optional ConfigMaps now log at debug level, while real read errors such as RBAC forbidden log at warn so they are easier to notice. Pull request #4113
- Added a RateProvider interface to pkg/currency so the exchange rate provider is configurable instead of tied to exchangerate-api.com. Pull request #4139
- Simplified CPU-throttling diagnostic query formatting with indexed arguments and added 55 regression cases across all 11 diagnostics. Pull request #4115
Why it matters
This week brings fixes that affect what people see in their cost and GPU data: output processing time is computed from a metric that actually measures decode duration, and GPU usage is detected when clusters report raw Kubernetes resource names. Release artifacts now come from the exact version tag, and genuine ConfigMap read errors are surfaced at warn instead of being buried among startup lines.
Changelog entry
- Fix output processing time calculation by not using vllm:request_time_per_output_token_seconds_sum as decode duration. Pull request #4154
- Fix GPU request, usage, and shared-GPU detection for raw Kubernetes resource names such as nvidia.com/gpu. Pull request #4134
- Build release image and SBOM from the exact vX.Y.Z tag for tag pushes and manual runs. Pull request #4100
- Log missing optional ConfigMaps at debug and real read errors at warn. Pull request #4113
- Add RateProvider interface to pkg/currency to make the exchange rate provider configurable. Pull request #4139
- Simplify CPU-throttling diagnostic query formatting with indexed arguments and add 55 regression cases across all 11 diagnostics. Pull request #4115
This week: fixed output processing time and GPU detection for raw Kubernetes resource names. Release images and SBOMs now build from the exact vX.Y.Z tag. Missing optional ConfigMaps log at debug, real errors at warn. Currency rate provider is now configurable.
This week in Opencost: output processing time no longer relies on a metric that does not measure decode duration, and GPU request, usage, and shared-GPU detection now works for sources reporting raw Kubernetes resource names like nvidia.com/gpu. Release images and SBOMs build from the exact vX.Y.Z tag. Missing optional ConfigMaps log at debug while real read errors log at warn. The currency converter gets a RateProvider interface so the exchange rate provider is configurable. The CPU-throttling diagnostic uses indexed arguments with 55 new regression cases across all 11 diagnostics.
Week of October 19, 2026
What shipped
- The allocation costs query now returns every allocation set instead of only the first, so narrow steps over a longer window no longer drop later days. Pull request #4111
- Node cost annotations are no longer written into the provider pricing cache, preventing overrides from leaking into cached node pricing for providers that return cached nodes. Pull request #4127
- The GCP billing pricing path now closes the HTTP response body and checks the status code, matching what the other cloud providers already did. Pull request #3941
- The missing AWS service key log line now logs at debug instead of info, so it no longer appears at default log level on IRSA and Pod Identity setups. Pull request #4083
- A warning is now logged every minute while a scrape is still outstanding, making stalled targets that pause metric collection visible. Pull request #4138
- Allocation, asset and efficiency queries now receive the request context, so work stops when the client times out or disconnects. Pull request #4110
- Changing settings now flushes the pricing cache by downloading pricing data again after the change is logged. Pull request #3604
- Contributor docs now include steps for testing Collector Source on kind, and dead paths were removed from the contributing guide. Pull request #4141
Why it matters
This week fixes several cases where results could be silently incomplete or stale: allocation queries dropped everything after the first set, request cancellation was ignored for allocation, asset and efficiency queries, and node cost annotations could be written into the provider pricing cache. Users also get a warning when a scrape hangs, and the GCP billing path now closes response bodies and checks status codes like the other providers. AWS users with IRSA or Pod Identity no longer see an unnecessary startup log line at default level.
Changelog entry
- Return every allocation set from allocation costs queries instead of only the first. Pull request #4111
- Stop writing node cost annotations into the provider pricing cache. Pull request #4127
- Close the GCP billing API response body and validate the HTTP status code. Pull request #3941
- Lower the AWS missing service key log to debug. Pull request #4083
- Log a warning every minute while a scrape is still outstanding. Pull request #4138
- Pass the request context to allocation, asset and efficiency queries. Pull request #4110
- Flush the pricing cache when settings change. Pull request #3604
- Add steps for testing Collector Source on kind to contributor docs. Pull request #4141
This week: allocation costs now return every allocation set, node cost annotations no longer leak into the provider pricing cache, GCP billing closes response bodies and checks status codes, and stalled scrapes log a warning.
This week in OpenCost: the allocation costs query now returns every allocation set instead of only the first, so steps smaller than the window no longer lose later days. Node cost annotations are no longer written into the provider pricing cache. The GCP billing pricing path now closes the HTTP response body and validates the status code, matching other providers. Allocation, asset and efficiency queries now receive the request context. A warning is logged every minute while a scrape is still outstanding, and the missing AWS service key log moved to debug so it no longer shows at default level. Settings changes now flush the pricing cache.
Week of October 12, 2026
What shipped
- Fixed parsing of UTC offset minutes in timeutil, so allocation and asset query windows use the correct timezone for positive, negative, and quarter-hour offsets. Pull request #4106
- Added support for custom GPU prices from node annotations, following the existing approach for CPU and RAM. Pull request #4114
- Added GBP, EUR, and JPY to the public pricing module. Pull request #4107
- Exported GetAssetKeyWithLabelConfig so callers can reproduce the asset key Insert would choose when keying results outside an AssetSet. Pull request #4094
- Ran the current Bingen version on KubeModel to properly include the shared library for streaming. Pull request #4108
Why it matters
This week the project fixed a timezone parsing bug that affected allocation and asset query windows, and added custom GPU pricing from node annotations so GPU costs can be configured like CPU and RAM. The pricing module now includes GBP, EUR, and JPY, improving pre-conciliation accuracy for users who switch to those currencies.
Changelog entry
- Fixed UTC offset minute parsing in timeutil to use the correct unit, affecting allocation and asset query windows. Pull request #4106
- Added custom GPU price extraction from node annotations. Pull request #4114
- Added GBP, EUR, and JPY to the public pricing module. Pull request #4107
- Released modules/pricing/public/v1.0.9 with GBP, EUR, and JPY support. Release
- Exported GetAssetKeyWithLabelConfig to allow keying with a label config. Pull request #4094
- Updated Bingen on KubeModel to include the shared library for streaming. Pull request #4108
This week: fixed UTC offset parsing for correct timezone windows, added custom GPU pricing from node annotations, and added GBP/EUR/JPY to the public pricing module. Also exported GetAssetKeyWithLabelConfig and improved KubeModel streaming.
This week in OpenCost: we fixed a timeutil bug so UTC offsets like +05:30 are parsed with the correct unit, keeping allocation and asset query windows in the right timezone. We added support for custom GPU prices from node annotations, following the same pattern as CPU and RAM. The public pricing module now includes GBP, EUR, and JPY, which should improve pre-conciliation accuracy for those currencies. We also exported GetAssetKeyWithLabelConfig for callers keying results outside an AssetSet, and ran the current Bingen version on KubeModel to properly include the shared library for streaming.
Week of October 5, 2026
What shipped
- KubeModel now exports all objects by default instead of only a subset of v1 objects, and the default provider is set to Custom rather than left blank. Pull request #4099
- Logs can now be excluded based on substring patterns, and the exclusion substrings can be changed via an API without restarting the pod. Pull request #4076
- Verbose and redundant log output during Allocation queries was reduced. Pull request #3985
- GCP auth secret and reserved instance log noise was reduced for repeated pricing download failures. Pull request #4084
- GCP pricing was split into subcategories for the n2 and n4 families. Pull request #4098
- The Prometheus model-label diagnostic queries in the inference cost guide were URL-encoded so the documented curl commands no longer produce invalid queries and HTTP 400 responses. Pull request #4088
- The public pricing module was updated for cluster management and now adds a default non-region Azure price for services. Pull request #4096
Why it matters
This week reduces log noise in Allocation queries and on GCP pricing downloads, which helps operators who see those messages repeatedly. It also changes KubeModel to export all objects by default, while adding log exclusion patterns that can be changed by API without restarting a pod.
Changelog entry
- KubeModel now exports all objects by default and defaults the provider to Custom. Pull request #4099
- Logs can be excluded by substring patterns, changeable via API without a pod restart. Pull request #4076
- Reduced verbose and redundant logs during Allocation queries. Pull request #3985
- Reduced GCP auth secret and reserved instance log noise on pricing downloads. Pull request #4084
- GCP pricing split into subcategories for n2 and n4 families. Pull request #4098
- Released pricing module v1.0.8 with updated GCP pricing. Release
- Updated public pricing module for cluster management and added a default non-region Azure price for services. Pull request #4096
- Released pricing module v1.0.7 with the cluster management update. Release
- URL-encoded Prometheus model-label diagnostic queries in the inference cost guide. Pull request #4088
KubeModel now exports all objects by default and defaults the provider to Custom. Logs can be excluded by pattern and updated via API without a pod restart. Allocation and GCP log noise is reduced, GCP pricing is split for n2/n4, and the inference cost guide queries are fixed.
Several changes landed this week. KubeModel now exports all objects by default instead of only a subset of v1 objects, and the default provider is Custom rather than blank. Logs can be excluded based on substring patterns, and those exclusions can be changed through an API without restarting a pod. Verbose log output during Allocation queries was reduced, and GCP auth secret and reserved instance log noise was reduced. GCP pricing is split into subcategories for the n2 and n4 families, and the public pricing module adds a default non-region Azure price for services. The documented Prometheus model-label diagnostic queries in the inference cost guide are now URL-encoded so they no longer return HTTP 400.
Week of September 28, 2026
What shipped
- Released modules/pricing/public/v1.0.6 which swaps defaults of cluster management for kubemodel pricing modules. Release
- Released modules/pricing/public/v1.0.5 which updates Azure public pricing defaults for cluster management and load balancers. Release
- Fixed bugs in KubeModel generation including provider support lists and PV disk asset provider IDs for GCP and Azure. Pull request #4089
- Inverted defaults to 0 with AWS/GCP cluster management costs set to 0.1 in kubemodel pricing modules. Pull request #4090
- Updated Azure public pricing to match GCP/AWS defaults for cluster management and load balancer costs. Pull request #4085
Why it matters
Two pricing module releases fix KubeModel generation bugs and align Azure pricing defaults with other cloud providers. These updates improve consistency across cloud provider pricing calculations.
Changelog entry
- Swap defaults of cluster management for kubemodel pricing modules - inverts default to 0 and sets AWS/GCP cluster management costs to 0.1 Release
- Update Azure public pricing defaults for cluster management and load balancers Release
- Fix KubeModel generation bugs: update provider support list and fix PV disk asset provider IDs for GCP and Azure Pull request #4089
Opencost v1.0.5 and v1.0.6 released. Updates to Azure pricing defaults and KubeModel cluster management costs, plus bug fixes for provider compatibility.
We released Opencost v1.0.5 and v1.0.6 this week. The updates include revised Azure public pricing to align cluster management and load balancer defaults with AWS and GCP, along with improvements to KubeModel generation. Bug fixes address provider support lists and disk asset provider IDs across cloud platforms.
Week of September 14, 2026
What shipped
- V1.121.3 released with allocation autocomplete label alias support, local disk pricing updates, GCP cloud configuration fixes, AWS marketplace fee inclusion, and IBM Cloud integration. Release
- IBM Cloud Usage Reports cloud cost integration added, enabling Kubecost to ingest IBM Cloud billing through the Usage Reports API. Pull request #4044
- GPU pricing added to GCP public pricing data. Pull request #4052
- Node labels concurrent modification issue fixed by cloning contents before pricing key generation. Pull request #4050
- Basic pricing split into on-demand and spot variants on the module side instead of aggregator side. Pull request #4051
- GCP cloud configuration now properly reads the ExcludePartitionTime field for BigQuery queries. Pull request #4038
- Allocation autocomplete now supports label aliases for Department, Owner, Environment, Team, and Product. Pull request #4045
Why it matters
V1.121.3 brings expanded cloud provider support with new IBM Cloud integration and improved cost accuracy through AWS marketplace fee inclusion and GCP configuration fixes. The release also enhances core functionality with concurrent modification fixes and better pricing data organization across modules.
Changelog entry
- Allocation autocomplete: Support label aliases (Department, Owner, Environment, Team, Product) Release
- Change local disk pricing in Kubemodel Release
- Read GCP cloud configuration field ExcludePartitionTime Release
- Fix AWS to include marketplace fees in Athena CUR integration Release
- Add IBM Cloud Usage Reports cloud cost integration Release
- Add GPU to GCP public pricing Release
- Fix Node.Labels concurrent modification by cloning before changes Pull request #4050
- Split basic pricing into on-demand and spot on module side Pull request #4051
OpenCost v1.121.3 is out. New IBM Cloud support, AWS marketplace fees, GCP GPU pricing, improved label aliases, and critical fixes for concurrent modifications and pricing calculations.
OpenCost v1.121.3 release adds IBM Cloud Usage Reports integration for cloud cost tracking, includes AWS marketplace fees in cost calculations, introduces GCP GPU pricing, supports label aliases in allocation autocomplete, and fixes concurrent modification issues in node label handling. Several module-level improvements optimize pricing data organization.