Skip to main content
​

Product

  • BYOK now includes 250,000 requests per workspace each UTC month without a Phaseo fee, then charges 2.5% of the provider-equivalent cost; request-level BYOK metadata is kept for 90 days (#2056).
​

Product

  • Chat errors now say what went wrong and offer a next step, such as retry, sign in or pick another model; 402 errors link to Credits and phaseo/free (#2100).
  • The Chat composer stops growing at five lines and scrolls long prompts instead of covering the conversation (#2096).
  • Model pages open a provider’s details from a link such as ?provider=openai, and the provider dialog shows a copyable provider-qualified model ID; see Provider-qualified model IDs (#2095, #2052).
  • Model search shows only models you can route through the gateway until you change the status filter; see Check Gateway Reliability and Status (#2057).
  • The public Apps rankings group each website into one entry with a readable URL, such as /apps/phaseo-chat (#2087).
  • The Trust Centre adds a security whitepaper, a subprocessor list and a draft DPA, and account deletion now removes data from every active Phaseo store within 30 days (#2058).

API & SDKs

  • The CLI asks before installing a missing coding agent and shows the exact install command; pass --skip-install for tools you manage yourself; see Coding Agents (#2066).
  • The management API manages provider credentials at /v1/byok and workspace budgets at /v1/budgets, and every SDK is regenerated to match; see List provider credentials (#2086).
​

Product

  • Files can be dragged into Chat to attach them, including HEIC and AVIF images (#2050).

API & SDKs

  • Requests accept labels in the X-Phaseo-Metadata header, and Logs and Activity can filter requests and spend by label; see Request Labels (#2049).
​

Product

  • Providers can claim their Phaseo profile with a domain proof and submit a model catalog feed for review from provider settings; see Publish a provider catalog (#2037).
​

Product

  • Request details list every provider the router ranked or excluded, with each score factor and whether the provider was selected, attempted or deranked; see Routing and Fallbacks (#2011, #2015).
  • Chat marks day changes, gaps of six hours or more, and model switches, showing the logos of the new models (#1992).
  • Model pages show performance and uptime for models with few requests; before, these sections stayed empty below 20 requests (#1997).
  • The release calendar shows a month overview with a selected-day agenda on small screens, and “On this day” uses the full release history (#2019).

API & SDKs

  • Balanced routing now weighs reliability, latency, tail latency, throughput, price and token fit, and provider uptime no longer counts 400, 403, 413 or 429 responses as failures; see Routing and Fallbacks (#2011).
​

Product

  • The Rankings page adds a weekly Return Rate section showing which models workspaces keep coming back to, with confidence ranges (#1961).
  • Global search ranks labs, providers and models by relevance once you type, so exact matches such as Nvidia appear first (#1974).
  • Model cards mark retired and deprecated models with End of life or Deprecated badges (#1977).
  • Provider privacy details show Data policy, ZDR and processing locations instead of separate training and retention labels (#1954).
  • Clicking the Phaseo wordmark opens a brand menu that copies or downloads the logo and wordmark as SVG or PNG, in light or dark versions (#1734, #1968, #1983).

API & SDKs

  • POST /v1/music/generate returns a Phaseo request ID, and GET /v1/music/generate/{music_id} returns results in one shape for every provider; see Generate music (#1941, #1952).
  • Failed requests forward the provider’s Retry-After, and BYOK requests also return X-Phaseo-Upstream-RateLimit-* quota headers; see Rate Limits (#1960).

Models

GLM 5.3 Flash model release artwork
​

Product

  • Each workspace alert can go to its own notification destinations, so low-balance and model deprecation alerts can reach different channels; see Workspace Notifications (#1830).
  • The Priority service tier is now shown as Fast on model and pricing pages; requests that send priority keep working; see Service Tiers (#1931).
  • Compare keeps the current models when you add another, and shows provider logos and a tier choice for providers with several plans (#1826).
​

Product

  • Workspace owners can buy an Enterprise plan for 100 to 100,000 members, with SSO, SCIM 2.0, directory and department settings and no committed model spend (#1754, #1743, #1749).
  • Workspaces can send low-balance, billing and model deprecation alerts to email, Discord, Slack, Microsoft Teams or a custom webhook from Workspace settings -> Notifications; see Workspace Notifications (#1796).
  • The Trust Centre at /trust describes Phaseo’s security practices, data handling, service providers and compliance status (#1811, #1816).
  • Organisation pages are now Labs pages, with aggregated performance and recent model cards (#1807).
  • Compare opens straight to the comparison view instead of a preset picker (#1803).
  • BYOK settings count remaining free requests against the 1,000,000 monthly allowance; before, they used the retired 100,000 allowance (#1595).

API & SDKs

  • Requests can classify an app with the X-App-Categories header, which the TypeScript SDK also supports; see App Attribution (#1799).
  • The :nitro, :cheap and :fast model suffixes now set the routing mode even when request fields, presets or workspace defaults ask for another; see Routing and Fallbacks (#1814).
  • provider.quantizations now restricts routing to offers with a matching quantization and returns diagnostics when none match; see Choosing a Quantization Strategy (#1810).
​

Product

  • The Providers page filters by BYOK support, training policy, data retention and datacenter region, and labels external providers (#1755, #1758, #1748).
​

Product

  • Model pages show pricing history by provider and service tier, with date ranges, list or effective prices, and CSV export (#1557).
​

Product

  • Selecting text in a Chat answer shows Explain, Improve, Shorten, Change tone and Fix grammar actions that place a prompt in the composer (#1619).
  • Chat keeps an unsent prompt through sign-in and returns you to the same chat and model (#1728).
  • Logs show each request’s Source, such as Direct HTTP, an SDK by language or a coding agent, separately from its App (#1723).
  • Logs open sessions and async jobs in side sheets with a request timeline, previous and next buttons, and an Investigate lookup by request ID (#1724).
  • Connected Apps group permissions by resource, and owners can remove individual scopes without revoking the whole app (#1585).
  • Mobile Models filters show active filters as removable chips and keep every filter when you switch between cards and table (#1590).
  • The Tools page adds a content provenance checker for image and audio uploads up to 20 MB (#1529).
  • BYOK keys can be saved again, and the key list updates right after a change; before, saving a key failed (#1591).
  • Logs and Usage Overview open an accessible workspace when the saved one is no longer available, and creating a workspace works again (#1589, #1592, #1593).
​

API & SDKs

  • Public presets referenced as @publisher/slug resolve again, and private presets work only with API keys created by their owner; see Presets (#1707, #1706).
​

Product

  • Privacy settings now live under Workspace -> Privacy and apply to Chat, API keys, SDK traffic and apps across the workspace (#1676, #1678).

API & SDKs

  • phaseo <agent> sets up Codex, Claude Code, OpenCode, Hermes, Pi, Prime Agent, DeepSeek Harness or OpenClaw with its own Phaseo key in one command; see Coding Agents (#1685, #1662).
  • Streamed POST /v1/messages responses send text, reasoning and tool-argument deltas as they arrive; before, they arrived in one burst at the end; see Create a message (#1679, #1681).
  • Providers whose support for a request parameter is unknown stay eligible for routing; only providers known not to support it are excluded (#1667).
​

API & SDKs

  • DeepSeek V4 Pro 0813 and DeepSeek V4 Flash 0731 accept reasoning_effort, and minimal is sent as low (#1651, #1652, #1654).
​

Product

  • Chat sends one session ID per conversation, so a conversation’s requests are grouped together in Logs (#1599).
  • Model pages replace Quickstart with a Use This Model panel that covers key setup, code examples by language, supported endpoints and parameters (#1637).
​

Product

  • Settings use one sidebar with an Account and Workspace switcher in place of separate navigation patterns (#1509).
  • Usage Overview opens for signed-in users on mobile; before, it could send them back to the sign-in page (#1556).

API & SDKs

  • Gemma 4 requests accept reasoning_effort and leave room for the final answer; before, small output budgets could return empty 502 responses (#1550).
​

Product

  • The OAuth consent screen groups requested permissions by resource and summarises read, write and delete access (#1538).

API & SDKs

  • Breaking: GET /v1/models returns a new Phaseo catalogue shape with lifecycle, modalities, limits, capabilities, pricing and provider offers; update code that reads the old fields; see List models (#1531).
  • The remote MCP server supports protocol version 2026-07-28, and OAuth clients can register with a Client ID Metadata Document URL; see CLI and MCP (#1531).
  • @phaseo/cli 0.2.1 adds phaseo update and phaseo doctor, which find the package manager that installed the CLI and report PATH conflicts; see Phaseo CLI (#1534, #1535).
  • phaseo integrations setup configures Codex, Claude Code or OpenCode to use Phaseo without writing API keys into their config files; see Coding Agents (#1510, #1545).
​

Product

  • Model pages add a section navigator on mobile, and old model sub-pages redirect to the matching section of the model page (#1508).
​

Product

  • Model pages chart the share of input tokens served from cache, over time and by provider (#1499).
​
​

Product

  • Model pages chart TTFT, output speed, TPOT, ITL and Phaseo overhead from P01 to P99, split by streaming and input length (#1344).
  • Presets save edits as drafts and publish them as numbered versions that never change, and marketplace forks pin to a source version; see Presets (#1419).
  • Broadcast OTLP destinations export gateway traces in the OpenTelemetry GenAI format; prompt and completion content is sent only if you opt in; see Broadcast (#1355).
  • Chat shows animated orbs while a model is working or composing a reply, and a new acknowledgements page lists the open-source projects Phaseo relies on (#1459).

API & SDKs

  • GET /v1/models/{author}/{slug}/endpoints lists each provider’s endpoints, supported parameters and pricing for a model; see List model endpoint capabilities (#1274).
  • @phaseo/ai-sdk-provider supports AI SDK 7, including tools, reasoning, embeddings, images, transcription and speech; AI SDK 6 stays on the 1.x line; see AI SDK (#1336).
  • POST /v1/feedback and POST /v1/events record ratings and events against requests, sessions and presets, using new feedback:read and feedback:write key scopes; see Collect Feedback for Presets (#1033).
​

Product

  • Request details list every provider attempt, including retries and failovers, with its finish reason and cost (#1276).
  • Side-by-side Chat comparisons use up to 96rem on wide screens and share the space between the selected models (#1365).

API & SDKs

  • Server tools add Subagent, Fusion and model search, enforce max_uses and max_tool_calls, and fetch up to 100,000 characters of web content; see Server Tools (#1447).
  • Requests call a preset by setting model to @slug, or @username/slug for a public preset; there is no separate preset field; see Presets (#1418).
  • POST /v1/audio/transcriptions accepts OpenAI’s diarized response formats, chunking_strategy and known-speaker fields; see Create a transcription (#1451).
​

Product

  • Usage adds a Geography tab, Rankings show 30-day usage by country, and new users choose their country during sign-up (#1425).
  • Global search lists your workspaces, and selecting one switches to it and opens its settings (#1417, #1424).
  • Pressing Send in Chat with no model selected opens the model picker; before, the button did nothing (#1421, #1423).
​

Product

  • Temporary chats show their messages with a dashed outline, and switching temporary mode on or off keeps your draft (#1399, #1362).
Last modified on October 28, 2026