Features
- Dropdown menus and tab selectors now use Base UI render composition consistently across chat, settings, usage, rankings, and data screens (PR #938).
- Amazon Bedrock provider coverage was staged for SpaceXAI Grok 4.3 with Bedrock Mantle model id
xai.grok-4.3, standard pricing, and currentus-west-2in-region availability. The provider-model capability is currently disabled pending setup. AWS currently lists Grok 4.3 as the only SpaceXAI model on Bedrock, and does not list Geo or Global inference for it. - Amazon Bedrock OpenAI frontier model coverage was checked; AWS still documents GPT-5.5 and GPT-5.4 as regional/Geo in-region offerings, with Global cross-region pricing marked as coming soon.
Model Releases
Features
- Chat now surfaces clearer timing and tool metadata, audio attachments, model-picker focus handling, tag management, and account credit display (PR #922).
- Header search now lazy-loads a compact cached index only when opened, reducing normal dashboard payloads (PR #920).
- Chat streaming, tool-call continuation, and model catalogue cache refresh paths were hardened for more reliable server-tool turns and model pages (PR #919, PR #918).
- API key reveal flows, internal navigation, and new-model Discord notifications were tightened for cleaner settings, admin, and release workflows (PR #812, PR #925, PR #927).
- SiliconFlow provider coverage was added for Tencent Hy3, including the free route and current pricing.
- xAI catalogue and provider labels were renamed to SpaceXAI across model pages, provider records, aliases, logos, and generated SDK/OpenAPI surfaces.
- Aion 1.0 and Aion 1.0 Mini were marked deprecated and retired on July 4, 2026 after Aion Labs removed them from the live model endpoint.
Model Releases
- Aion Labs: Aion 3.0
- Aion Labs: Aion 3.0 Mini
- Meta: Muse Image
- Meta: Muse Video
- Sakana: Fugu
- Sakana: Fugu Ultra
- Tencent: Hy3
- Microsoft MAI and Phi catalog entries
Features
- AI Stats is now Phaseo across the product, documentation, SDK examples, and public site. Read the migration announcement in AI Stats is becoming Phaseo.
- New developer-facing examples use Phaseo naming and
PHASEO_*configuration while existing deployments can keep their current environment-variable names. - The blog and announcement surfaces were refreshed for the rebrand, including the new migration announcement layout and launch imagery.
- Local, preview, and production builds now use distinct favicons so active environments are easier to tell apart while testing.
Model Releases
Provider Coverage
- Amazon Bedrock Mantle was split into its own provider catalogue with model coverage and pricing for Anthropic, Google, MiniMax, MoonshotAI, NVIDIA, OpenAI, Qwen, SpaceXAI, and Z.AI models.
- AtlasCloud and Novita provider coverage was added for Tencent Hy3, including free-route pricing where available.
- Groq Qwen 3.6 gateway routing was disabled until provider pricing is available.
Pricing
- DeepSeek V4 Flash and DeepSeek V4 Pro pricing now supports time-windowed provider rates.
Features
- Server-tool inputs, monitor history links, workspace credit cache invalidation, and pricing edge cases were hardened across gateway and web surfaces (PR #900).
- Chat gateway target resolution now fails closed in production when the configured gateway URL is missing, while preserving local-development overrides (PR #899).
- Anthropic and Bedrock request shaping now uses adaptive thinking controls for Claude Sonnet 5 and newer Opus 4.x models (PR #901).
- Server-tool handling was hardened for datetime, web fetch, web search, image generation, and fusion workflows.
- Monitor links now apply stricter URL safety checks.
- GMICloud MiniMax M3 cached-read pricing and Google Vertex Gemini 3.1 Flash-Lite Image pricing validation were tightened.
Features
- Individual model pages were overhauled with clearer lifecycle notices, restored routing details, and better handoffs from model pages into gateway workflows (PR #888).
- Retired and replaced model pages now make status, alternatives, and historical context easier to scan without burying the current model link (PR #888).
- Chat tags, tool-call markers, API target controls, and mobile viewport handling were restored or tightened across the chat experience (PR #894, PR #895, PR #891).
Features
- Core interactive UI primitives moved onto Base UI, improving accessibility and behaviour consistency across filters, menus, dialogs, and model controls (PR #854).
- Chat and settings flows received follow-up polish after the Base UI migration, including steadier model pickers and cleaner menu spacing (PR #853).
Model Releases
Features
- Pricing calculators now support time-windowed pricing and side-by-side model comparisons, making burst, realtime, and route-specific pricing easier to reason about (PR #864).
- The pricing calculator layout was simplified so model, provider, and usage assumptions are easier to adjust before comparing costs (PR #864).
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Features
- Pricing selectors now present Qwen 3.7 Max variants more clearly, reducing ambiguity between preview, dated, and current model routes (PR #418).
- Anthropic-on-AWS provider branding was refreshed so Bedrock and Claude Platform routes are easier to distinguish in provider lists (PR #415).
Model Releases
SDK Releases
- The TypeScript Agent SDK launch added a dedicated package for building agentic applications on top of Phaseo Gateway (PR #495). See
@phaseo/agent-sdk.
Model Releases
Features
- Provider activation states and region filters were tightened so unavailable routes are easier to identify before a request is made (PR #495).
- Free-router behaviour was made more predictable across pricing, provider selection, and gateway routing surfaces (PR #495).
SDK Releases
- Agent SDK documentation and examples shipped with the first public workflow guidance for durable loops, parallel tools, and agent orchestration (PR #495).
Model Releases
Features
- Provider availability and pricing summaries were expanded so model pages show clearer route, tier, and cost differences (PR #474).
- Realtime minute pricing became visible in the product, making audio and live-session models easier to compare against token-priced models (PR #477).
Model Releases
SDK Releases
- TypeScript SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- Python SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- Go SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- C# SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- Java SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- PHP SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
- Ruby SDK 2.0.4: generated OpenAPI and model-surface updates for the latest gateway schema.
SDK Releases
- SDK
2.0.2shipped for TypeScript, Python, Go, C#, Java, PHP, and Ruby. - This release separated runtime request model IDs from generated callable-helper constants, so newly released models can be used before the next SDK publish.
- Generated helpers now track the current callable-on-gateway snapshot, reducing noisy SDK churn from catalog-only model updates.
Model Releases
Features
- SDK
2.0.0shipped with regenerated clients and broader parity against the current gateway contract.
Model Releases
Features
- The announcements hub, workspace management surfaces, and related membership flows shipped.
Model Releases
Features
- SDK
1.2.0synced model availability, generated types, and docs with the latest catalog changes.
Features
- Mobile model pages, pricing layouts, and free-route visibility were improved, and the Pay As You Go top-up fee dropped to 5%.
Features
- New language SDKs shipped, TypeScript and Python SDKs were overhauled, and chat rooms went multimodal.
- Foundry launched, LLM Council became the first Foundry project, and SDK
1.1.0aligned model availability and lifecycle warnings with the API.
Model Releases
Features
- The models display page was overhauled for clearer availability, metadata hierarchy, and support context.
Model Releases
Features
- A broader brand refresh shipped across the website and docs.
- Alibaba Cloud provider coverage landed across Phaseo, gateway routes, and documentation.
Model Releases
Features
- GPT 5.4 and GPT 5.4 Pro landed across discovery, gateway, and docs.
Model Releases
Features
- Gemini 3.1 Flash Lite Preview landed across product surfaces, gateway routing, and docs.
Model Releases
Features
- Responses API assistant
phasesupport landed across request handling, response payloads, and OpenAPI docs.
Model Releases
Features
- Gemini 3.1 Flash Image Preview support landed across discovery, gateway compatibility, and docs.
Model Releases
Features
- Gemini 3.1 Pro Preview coverage landed across discovery, pricing, gateway support, and docs.
Model Releases
Features
- Claude Sonnet 4.6 coverage landed across discovery, gateway compatibility, and docs.
Model Releases
Model Releases
Features
- The gateway was overhauled again, including Anthropic-compatible
/messagessupport. - Database architecture, website UX, playground tooling, and SDK direction were all refreshed for broader expansion.
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Features
- Early alpha SDK releases landed for Python and TypeScript or JavaScript.
Model Releases
Features
- OpenAI-compatible
/responsessupport, aligned SDKs, and gateway quickstart or pricing page updates shipped. - Docs expanded ahead of the SDK launch, and nightly CI started automatically refreshing model statuses.
Model Releases
Model Releases
Model Releases
Model Releases
Model Releases
Features
- Phaseo relaunched with a refreshed brand, a broader gateway, and much wider data coverage.
- Model availability, benchmark pages, provider pages, and the latest-updates surfaces were overhauled.
- The unified gateway entered alpha, and the first Python and TypeScript or JavaScript SDKs launched alongside it.