AI Model Release Dashboard

Unified model combining text, vision, reasoning, and coding. MoE 119B total / 6B active params. 256K context. Apache 2.0.

Most capable small OpenAI model. 2x faster than GPT-5.4. Strong on coding, reasoning, and computer use. Free and Go users.

Smallest and cheapest OpenAI model. Optimized for classification, data extraction, ranking, and coding subagents.

Most capable frontier model for professional work. Native computer use, 1M context, consolidated coding+reasoning. 33% fewer hallucinations vs GPT-5.2.

Ultra-efficient Gemini 3.1 model for developers. Fast and cost-effective for high-volume API workloads.

Smoother, more helpful everyday conversations. 400K context. 26.8% fewer hallucinations than GPT-5.2 with web search. Reduces unnecessary refusals and moralizing preambles.

Faster image generation with 4K resolution, improved text rendering, and multi-language support. #1 on Artificial Analysis Image Arena at launch. Default in Gemini app, Search, and Lens across 141 countries.

Advanced reasoning over complex problems. 77.1% on ARC-AGI-2. Available via Gemini API, Vertex AI, and NotebookLM.

Default model for free and Pro users. 1M token context (beta). Improved computer use, coding, and large-scale data processing.

1M token context window. 14.5-hour task completion horizon. Top Finance Agent benchmark. Stronger coding and planning.

Most capable agentic coding model to date. 25% faster than GPT-5.2-Codex. First OpenAI model instrumental in creating itself. State-of-the-art on SWE-Bench Pro and Terminal-Bench.

Agentic coding model optimized for long-horizon task completion. Context compaction for multi-window workflows. 56.4% on SWE-Bench Pro, 64.0% on Terminal-Bench 2.0.

Efficient Gemini 3 model. Includes Gemini 3 Pro reasoning in a faster, cheaper package. Default model in the Gemini app.

Next upgrade to the GPT-5 series. Smarter and more useful for work and learning. 400K context, 128K output. Available in Instant, Thinking, and Pro variants.

Extended-thinking version of Gemini 3 Pro. Rolled out to Ultra subscribers. Designed for the hardest reasoning tasks requiring longer deliberation.

MoE architecture: 41B active / 675B total parameters. Multimodal, 256K context. Apache 2.0. #2 on LMArena OSS leaderboard.

Family of dense small models in 3 sizes × 3 variants (Base/Instruct/Reasoning). Runs on 4GB VRAM. Apache 2.0 license.

Most robustly aligned Claude model. Reclaimed coding crown with 80.9% SWE-bench Verified. Significantly reduced pricing.

Best multimodal understanding model. Most powerful agentic and coding model from Google. Replaces Gemini 2.5 Pro.

Warmer personality, improved instruction-following. Adds shopping research and 8 personality options. Available in Instant, Thinking, and Pro variants. Retired March 2026.

Fast and affordable Claude 4 family model. Targets latency-sensitive applications. Available on Claude.ai and mobile.

Best coding model at launch with 77.2% on SWE-bench Verified. 30+ hour autonomous task horizon. Launched alongside Claude Agent SDK and Claude Code 2.0.

Dedicated image generation model from Google. Better instruction following and text rendering than previous image models.

Unifies reasoning and standard model capabilities. Automatically switches between fast and deep thinking modes based on task needs.

Drop-in replacement for Opus 4 with superior performance. Handles complex multi-step agentic tasks with more rigor. Improved precision for real-world coding workflows.

Next generation Claude model. Enhanced coding, reasoning, and agentic task performance.

Most intelligent Claude model. Designed for complex, long-horizon tasks and deep reasoning.

24B parameter open-weight model (Apache 2.0). Improved instruction following and function calling over v3.1. 84.78% on Mistral internal benchmark.

Efficient version of Gemini 2.5 (GA). Controllable thinking budget. Best price-performance in Gemini family.

Ultra-efficient Gemini 2.5 variant optimized for speed and cost. Lowest latency in the Gemini 2.5 family.

Extended-thinking version of o3. Thinks longer for more reliable responses on the hardest tasks. ChatGPT Pro and API.

Mistral's first reasoning model with chain-of-thought capabilities. 24B parameters, open-source under Apache 2.0. Released alongside Magistral Medium.

Flagship reasoning model. Top ARC-AGI scores. Available in API with high/medium/low reasoning effort settings.

Cost-efficient reasoning model with vision. Faster than o3, strong coding performance. Tool use support.

Specialized for coding and instruction-following. Stronger than GPT-4o at precise web dev tasks. API-only release.

Fast, capable small model. Replaces GPT-4o mini for paid users. Significant improvements in coding and instruction-following.

MoE architecture models: Scout (17Bx16E) and Maverick (17Bx128E). Native multimodal, 10M context window.

State-of-the-art reasoning model. 1M context, native audio/video/image understanding. Top LMArena scores.

Updated small model with vision capabilities added. 24B parameters, 128K context. Apache 2.0 license.

Largest and most capable GPT model. Improved emotional intelligence and world knowledge. 128K context.

First hybrid reasoning model. Switchable extended thinking mode. Top scores on SWE-bench Verified (70.3%).

Specialized for Arabic and South Asian languages. Strong Middle East/North Africa cultural context understanding.

Generally available release. Upgraded image generation, improved multilingual audio. 1M context. Free tier available.

Most capable Gemini 2.0 model for complex tasks. 2M context window. Code execution and Google Search grounding.

Cost-efficient small reasoning model. Three effort levels (low/medium/high). Outperforms o1 on AIME and SWE-bench.

24B parameter model with 81% MMLU. Apache 2.0 license. Best-in-class small model for latency-sensitive tasks.

Updated code generation model. 256K context window, supports 80+ programming languages. Available via API.

First Gemini 2.0 model. Multimodal I/O, native tool use, 1M context. Faster than 1.5 Pro. Agentic capabilities.

Updated 70B model with performance matching Llama 3.1 405B. More compute-efficient. Open weights.

Full reasoning model release. Includes vision support, 200K context. Available in ChatGPT and API.

124B parameter multimodal model. Strong document understanding, charts, and scientific diagrams. Available via API.

Fastest Claude model. Matches Claude 3 Opus on many benchmarks. 200K context. API and Bedrock availability.

Upgraded Sonnet with improved coding (SWE-bench 49%), vision, and computer use capability (beta).

Smallest Gemini model for high-frequency, low-latency tasks. 1M token context. Generally available via API.

Multimodal models (11B, 90B) with vision. Lightweight text models (1B, 3B) for edge deployment. Open weights.

Early access reasoning model with extended chain-of-thought. Designed for complex science, coding, and math tasks.

Smaller, faster reasoning model. Cost-effective for coding and STEM tasks without broad world knowledge needs.

First multimodal model from Mistral. 12B parameters with vision capabilities. Apache 2.0 license.

128K context, strong coding and multilingual capabilities. 123B parameters. Apache 2.0 license.

Open-weights models: 8B, 70B, 405B. 128K context window. First open model competitive with frontier closed models.

Affordable, fast frontier model. 128K context, supports vision. Lower cost alternative to GPT-4o.