Release Notes

OpenAI GPT-5.6

OpenAI's newest frontier generation, in three tiers. OpenAI cut the Luna and Terra prices on July 30, and those are the rates below.

• GPT-5.6 Sol: Highest intelligence tier ($5.00/$30.00 per 1M tokens)
• GPT-5.6 Terra: Balanced intelligence and cost ($2.00/$12.00, down 20%)
• GPT-5.6 Luna: Cost-sensitive workloads ($0.20/$1.20, down 80%)

Grok 4.5

xAI's new flagship for agentic software, engineering, and workflow tasks.

• 500K token context window ($2.00/$6.00 per 1M tokens)
• Strong function calling and structured outputs

Gemini 3.6 Flash & 3.5 Flash-Lite

Google's newest Flash tier, plus a cost-efficient 3.5-family option.

• Gemini 3.6 Flash: Latest stable Flash model ($1.50/$7.50)
• Gemini 3.5 Flash-Lite: High-volume, low-latency ($0.30/$2.50)
• Both support configurable thinking levels

Groq Qwen3.6 27B & GPT-OSS Safeguard 20B

Groq's replacements for the models it retired this month.

• Qwen3.6 27B: Multilingual reasoning, 131K context ($0.60/$3.00)
• GPT-OSS Safeguard 20B: Content moderation ($0.075/$0.30)

Cerebras Gemma 4 31B

Google's Gemma 4 31B on the Cerebras Wafer-Scale Engine at about 1,850 tokens per second.

• 131K context window on paid tiers ($0.99/$1.49)

Lower Prices on Existing Models

Provider price changes are now reflected in the model list and the cost calculator.

• Groq GPT-OSS 20B: $0.075/$0.30, down from $0.10/$0.50
• Groq GPT-OSS 120B output: $0.60, down from $0.75
• Gemini 2.5 Flash: $0.30/$2.50, corrected from $0.15/$0.60

Gemini 3 Pro & Gemini 3 Flash Removed

Google never promoted these two ids past preview, then shut the previews down. Any assistant on them fails. Move to Gemini 3.6 Flash or Gemini 3.5 Flash.

Groq Model Retirements

Groq decommissioned three more models. They are gone from the model picker. Existing assistants should move to Qwen3.6 27B or GPT-OSS 120B.

• Qwen3 32B and Llama 4 Scout shut down July 17, 2026
• Kimi K2 0905 shut down April 15, 2026

Groq Llama Retirement on August 16, 2026

Groq will shut down both remaining Llama models. They stay selectable until that date. Plan the move now.

• Llama 3.1 8B Instant: Replace with GPT-OSS 20B
• Llama 3.3 70B Versatile: Replace with GPT-OSS 120B or Qwen3.6 27B

Cerebras Public Endpoint Changes

Cerebras moved the Llama and Qwen families to Dedicated Endpoints, which this integration does not use. The GLM 4.7 model id was also wrong, so the model never resolved.

• Llama 3.1 8B, Llama 3.3 70B and Qwen 3 32B removed from the picker
• GLM 4.7 id corrected to zai-glm-4.7

OpenAI GPT-5.4 & GPT-5.5

OpenAI's newest flagship generation with frontier reasoning for coding and professional work.

• GPT-5.5: New flagship ($5.00/$30.00 per 1M tokens)
• GPT-5.4: More affordable frontier model ($2.50/$15.00)
• GPT-5.4 mini ($0.75/$4.50) & nano ($0.20/$1.25)
• GPT-5.3 Codex: Most capable agentic coding model ($1.75/$14.00)

xAI Grok

First-class Grok support via xAI's API, with the Grok 4.3 flagship and the Grok 4.20 reasoning/non-reasoning pair.

• Grok 4.3: Flagship with 1M context ($1.25/$2.50)
• Grok 4.20 Reasoning & Non-Reasoning variants
• Industry-leading non-hallucination rate and agentic tool calling

Gemini 3.5 Flash & Flash-Lite

Google's latest fast tier, plus the cost-efficient Gemini 2.5 Flash-Lite.

• Gemini 3.5 Flash: Frontier speed ($1.50/$9.00)
• Gemini 3.1 Flash-Lite: Low-latency, high-volume ($0.25/$1.50)
• Gemini 2.5 Flash-Lite: Most cost-efficient ($0.10/$0.40)

Groq Production Lineup

Current Groq-hosted models for ultra-fast inference.

• Llama 3.1 8B Instant: Cheapest on Groq ($0.05/$0.08)
• Qwen3 32B with switchable thinking modes ($0.29/$0.59)
• Kimi K2 0905: 1T-param agentic coding model ($1.00/$3.00)

Groq Model Retirements

Several older Groq-hosted models were decommissioned upstream and have been removed from the model picker. Existing assistants should migrate to the current Groq lineup above.

• Gemma2 9B, Mistral Saba 24B, Llama Guard 3 8B
• DeepSeek R1 Distill (Llama 70B & Qwen 32B)
• Qwen 2.5 32B / Coder 32B, Qwen QWQ 32B
• Llama 3 8B/70B (8192), Llama 4 Maverick

Gemini Legacy Models

Google has retired its older Gemini tiers. Move to Gemini 3.5 Flash, 3.1 Flash-Lite, or 2.5 Flash-Lite.

• Gemini 2.0 Flash & Flash-Lite shut down June 1, 2026
• Gemini 1.5 Flash removed from the current lineup

Cerebras Models

Ultra-fast inference powered by Wafer-Scale Engine technology.

• Llama 3.1 8B at $0.10/$0.10 per 1M tokens
• Llama 3.3 70B with function calling at $0.60/$0.60
• Qwen 3 32B supporting 29+ languages at $0.20/$0.20
• GPT-OSS 120B reasoning model at 3,000 TPS

Speed Improvements

Industry-leading throughput for high-volume applications with simple per-token pricing.

Gemini 3 Pro & Flash

Google's latest models with 1M token context and PhD-level reasoning (90.4% on GPQA Diamond).

• Gemini 3 Pro: $2.00/$12.00 per 1M tokens
• Gemini 3 Flash: 3x faster at $0.50/$3.00
• Native multimodal: text, images, audio, video

Agentic Workflows

Built-in support for real-time tool use, coding, and function calling.

GPT-5.2

OpenAI's flagship with 400K context, 128K output, and enhanced reasoning tokens.

• Pricing: $1.75 input / $14.00 output per 1M tokens
• Excels at agentic workflows and multi-step coding
• Enhanced structured document analysis

Full GPT-5 Suite

Complete model family from premium to cost-efficient variants.

• GPT-5: Premium flagship ($1.25/$10.00)
• GPT-5 mini: Balanced performance ($0.25/$2.00)
• GPT-5 nano: Fastest, most affordable ($0.05/$0.40)
• GPT-5.1 Codex: Agentic coding ($1.50/$12.00)

Open Source GPT Models

OpenAI's open-weight models available through Groq's ultra-fast infrastructure.

• GPT-OSS 20B: $0.10/$0.50 per 1M tokens
• GPT-OSS 120B: $0.15/$0.75 per 1M tokens
• Built-in browser search and code execution

Emotions Detection

Detect emotions in English conversations for deeper user understanding.

Sentiment Analytics

Track sentiment for both user input and AI responses with performance stats.

RAG Enhancements

AI-driven retrieval that chooses what to search and when, unified across all providers.

• Content deduplication and cost savings
• Context propagation while needed
• Detailed RAG analytics

New Models

LLaMA 4 Scout/Maverik via Groq, OpenAI GPT 4.1 family.

Provider Integration

One API for OpenAI, Gemini, and Groq with automatic failover.

Observability

Latency metrics, cost breakdowns, and cache savings tracking.