Skip to main content
​

API & SDKs

  • service_tier: "ultrafast" routes only to Ultrafast offers, starting with GPT-6 Astra, and fails instead of falling back to another tier; see Service Tiers (#2624, #2663).
​

API & SDKs

  • Text requests need at least $0.10 in available credits instead of $1, while video and batch submissions still need $1 (#2643).

Models

Claude Sonnet 5.5 model release artwork
​

Product

  • Global search tolerates typos and lists recent items, and the workspace switcher searches all your workspaces, showing the 50 most recently used first (#2572).
​

Product

  • Global search finds documentation pages by title, description and heading and opens them directly (#2526).
​

Product

  • Chat’s reasoning menu lists only the effort levels the selected model and provider support, and replaces a saved effort that no longer applies (#2496, #2503).

API & SDKs

  • The Responses API accepts reasoning.effort: "instant" for models that support it (#2503).
​

Product

  • Settings -> Preferences sets locale, date and time formats, time zone, number format, density, landing page and data masking, plus custom light and dark themes (#2452).
  • The Models and API Providers tables save column visibility, order, pinning and density, and show each model’s providers in one row (#2473).
  • Chat’s code samples add cURL, all nine SDK languages and Agent SDK starters, and can switch text requests between Responses, Chat Completions and Messages (#2468, #2469, #2487).
​

Product

  • Each Chat room has a Code button that turns the last request into TypeScript or Python, and each reply’s details link to its request log (#2443, #2453).
  • Model links shared in Discord show a card with the model’s name, lab and context window, and buttons to open, chat, compare or use the API (#2428, #2446).
  • Model provider tables show tier badges, group regional and specialised routes under Additional Routing Options, and hide external listings unless turned on in Filters (#2447, #2449).
​

Product

  • Chat adds OCR and Rerank rooms, and the Video and Realtime rooms are now open to everyone (#2440, #2242).
  • Chat adds a Decisions room where you pick the question type for each message and see answers with probabilities and scores (#2429).
  • Rankings show usage in all eight modality tabs, and Weekly Return Rate loads again instead of failing (#2439).

API & SDKs

  • Responses that contain only reasoning now return 200 with their length or stop finish reason and usage, instead of a 502 error (#2432, #2435).
​

API & SDKs

  • Beta: POST /v1/decisions answers typed noul, choice and score questions about application state, starting with TypeSafe Jev; see Make a decision (#2425).
​

Product

  • Logs tables save column visibility, order, pinning and density, and request details show how long Phaseo took before sending the request to the provider (#2415).

API & SDKs

  • Breaking: The TypeScript, Python, Go, C#, Java, PHP and Ruby SDKs 3.0.0 return the Phaseo model catalogue format from the models endpoint; update code that reads model fields (#1280).
  • The SDKs 3.0.0 add management methods for API and management keys, guardrails, budgets, presets, members, SSO and SCIM, notifications and Private Models (#1280, #2422).
  • Phaseo CLI 0.3.0 sets up coding agents such as OpenCode, and runs Cline, Kilo Code, oh-my-pi, Muse Code and MiniMax Code through Phaseo; see Phaseo CLI (#2325, #2336, #1280).
​

Product

  • Model pages and a new benchmark page show Epoch AI’s Capabilities Index score, rank, confidence interval and progress over time (#2408).

Models

​

Product

  • Chat, model and provider privacy rules now come only from the active workspace, including personal workspaces, and workspace privacy settings save again (#2400, #2403).
  • Turning on Auto Top-Up without two-factor authentication requires a typed confirmation, and existing setups without it show a warning (#2401).
  • Buying credits through Stripe checkout works again; before, checkout failed before reaching Stripe (#2405).

API & SDKs

  • The Terraform provider 0.1.0 manages API keys, guardrails, provider credentials, SCIM group mappings and observability destinations, and reads models, providers and credits (#2164, #2396).
  • API key and management key limits are counted separately, and only active, unexpired keys count toward them (#2399).
​

Product

  • Chat adds a navigation rail with a marker for each turn, so you can jump through long conversations (#2385).
  • The Chat model picker disables models your guardrails block, adds a Hide Unavailable filter, and Ctrl- or ⌘-click adds models to a comparison (#2389).
  • Changing your account country saves again; before, the change failed (#2392).
​

Product

  • Chat lets you choose Standard, Priority or Flex for each model and sends Priority and Flex as service_tier (#2357).
  • Provider cards on model pages chart uptime, latency and throughput hour by hour for the last 60 hours, marking hours with no data (#2361).
  • Chat shows reasoning as it streams, and each reply’s details show the gateway-measured throughput, output tokens and service tier (#2359, #2367).
  • Model deprecation alerts reach the workspace’s notification destinations, and deprecation warnings recommend a replacement model; see Model Deprecation and Retirement (#2136).

API & SDKs

  • Requests without service_tier, or with service_tier: "default", use Standard routes, and Flex-only offers serve only requests that ask for flex; see Service Tiers (#2356, #2354).
  • Requests that use gateway-managed tools stream each model turn’s text and tool calls as they arrive instead of all at once at the end (#2360, #2363).
​

Product

  • Settings -> API Keys can search by key name or by a pasted full key, which stays masked (#2344).
  • Model pages show provider latency, throughput and uptime from a provider’s first request instead of after 20 requests, and chart cache rate per provider (#2339, #2341).
​

Product

  • Model benchmark sections add Artificial Analysis charts for Intelligence, Coding, Agentic and evaluation cost, with model search and the other models dimmed (#2321, #2326).
  • Pricing history charts use each provider’s colour and draw service tiers as solid, dotted, dashed or dash-dot lines, matching the provider table (#2323).
​

Product

  • Model pages mark providers that need your own provider key, and Use This Model asks you to add one first when every route requires it (#2303).
​

Product

  • Model pages, benchmark pages and rankings lead with the Artificial Analysis Intelligence, Coding and Agentic indexes and evaluation cost, showing the tested configuration (#2264).
​

Product

  • Settings -> Billing & Credits shows an active Enterprise subscription beside usage credits and manages it through the Stripe portal (#2252).
  • Credit purchases link to their Stripe receipt instead of an invoice (#2256).
  • The API keys list hides expired keys by default, sorts enabled keys first, adds status filters and moves bulk actions to a floating bar (#2261).
  • Testing a key in Settings accepts current phaseo_v1_sk_ keys; before, newly created keys failed the test (#2251).
  • The settings sidebar has a filter that searches every Account and Workspace page, and global search suggests pages related to the one you’re on (#2259).
  • Model pages opened through an alias, such as openai/gpt-astra-latest, show that alias with an Alias label beside the default ID (#2226).

API & SDKs

  • SCIM provisioning from Okta works; Phaseo accepts the password, groups, id and meta attributes Okta sends and does not store them (#2240).
​

Product

  • Workspaces can add a dedicated or self-hosted OpenAI-compatible endpoint as a Private Model that only members see and call by its model ID; see Connect Dedicated Model Deployments (#2201).
  • Model pricing tables fold each provider’s service tiers under one row, and sorting a column lists every tier as its own row (#2198).
​

API & SDKs

  • The regional endpoints eu.api.phaseo.app and us.api.phaseo.app keep Chat Completions, Responses and Messages requests on EU or US provider routes; see Route text requests by region (#2165, #2171).
  • Function and custom tools accept async: true in Chat Completions, Responses and Messages; Phaseo sends it to OpenAI and drops it for other providers (#2193, #2194).
​

API & SDKs

  • phaseo:web_search accepts engine: "tinyfish" to search with TinyFish, with language hints, result pages and the shared domain filters; see Web Search Server Tool (#2124).
  • OAuth sign-in from native apps that use a loopback callback on a random port, such as coding agents connecting to the MCP server, no longer fails with invalid_client (#2129).
Last modified on October 28, 2026