New frontier AI model releases tracked by DemandSphere. Benchmarks, pricing, and capabilities from OpenAI, Anthropic, Google, xAI, Meta, DeepSeek, and more. https://www.demandsphere.com/research/demandsphere-radar/ai-frontier-model-tracker/releases/ en-us Sat, 16 May 2026 13:56:11 +0000 CC BY-NC 4.0 DemandSphere, Inc. 1M context. Three reasoning intensity levels. $1.25/$2.50 per 1M tokens. GA May 6 (beta Apr 17). Context: 2000K tokens. Pricing: $1.25/M input, $2.5/M output. https://piunikaweb.com/2026/04/17/xai-grok-4-3-beta-supergrok-heavy/ ds-tracker-grok43-2026-05-06 Wed, 06 May 2026 00:00:00 +0000 Reasoning Closed xAI New ChatGPT default, replacing GPT-5.3 Instant. 52.5% fewer hallucinations on high-stakes prompts (medicine, law, finance). 81.2 AIME 2025 (vs 65.4 prior). Vision supported. API alias: chat-latest. https://openai.com/index/gpt-5-5-instant/ ds-tracker-gpt-55i-2026-05-05 Tue, 05 May 2026 00:00:00 +0000 Reasoning Closed OpenAI 128B dense model. 256K context. Open weights (modified MIT). Strong coding and agentic capabilities. 77.6% SWE-Bench Verified. $1.50/$7.50 per 1M tokens. Context: 256K tokens. Pricing: $1.5/M input, $7.5/M output. Open weights. https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5 ds-tracker-mis-med35-2026-04-29 Wed, 29 Apr 2026 00:00:00 +0000 Dense Open Mistral AI 284B MoE / 13B active. MIT license. Same hybrid attention as V4-Pro. 88.1% GPQA Diamond, 91.6% LiveCodeBench. 1M context. $0.14/$0.28 per 1M tokens - 96% cheaper than GPT-5.4 output. Context: 1000K tokens. Pricing: $0.14/M input, $0.28/M output. Open weights. https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash ds-tracker-dsv4f-2026-04-24 Fri, 24 Apr 2026 00:00:00 +0000 Reasoning Open DeepSeek 1.6T MoE / 49B active. MIT license. Trained on 32T+ tokens. Hybrid Attention (CSA+HCA): 27% inference FLOPs and 10% KV cache vs V3 at 1M tokens. 90.1% GPQA Diamond, 80.6% SWE-bench Verified, 93.5% LiveCodeBench (SOTA), 37.7% HLE. Three reasoning modes (non-think, high, max). Context: 1000K tokens. Pricing: $1.74/M input, $3.48/M output. Open weights. https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro ds-tracker-dsv4p-2026-04-24 Fri, 24 Apr 2026 00:00:00 +0000 Reasoning Open DeepSeek Higher-accuracy variant of GPT-5.5 with extended compute. $30/$180 per 1M tokens. Context: 1000K tokens. Pricing: $30/M input, $180/M output. https://openai.com/index/introducing-gpt-5-5/ ds-tracker-gpt-55p-2026-04-24 Fri, 24 Apr 2026 00:00:00 +0000 Reasoning Closed OpenAI 295B MoE / 21B active. 87.2% GPQA Diamond, 74.4% SWE-bench Verified. Hybrid fast/slow thinking. 256K context. Tencent Hy Community License. Open weights. Context: 256K tokens. Open weights. https://huggingface.co/tencent/Hy3-preview ds-tracker-hy3-2026-04-23 Thu, 23 Apr 2026 00:00:00 +0000 Reasoning Open Tencent OpenAI's most capable model. Stronger coding, agentic tasks, and research. $5/$30 per 1M tokens. 1M context. Context: 1000K tokens. Pricing: $5/M input, $30/M output. https://openai.com/index/introducing-gpt-5-5/ ds-tracker-gpt-55-2026-04-23 Thu, 23 Apr 2026 00:00:00 +0000 Reasoning Closed OpenAI Dense 27B open-weight vision model. Apache 2.0. Hybrid Gated DeltaNet + Gated Attention. 87.8% GPQA Diamond, 77.2% SWE-bench, 83.9% LiveCodeBench v6. 262K native context (1M extended). Thinking preservation across turns. Context: 262K tokens. Open weights. https://huggingface.co/Qwen/Qwen3.6-27B ds-tracker-qw36-27b-2026-04-22 Wed, 22 Apr 2026 00:00:00 +0000 Reasoning Open Alibaba/Qwen Most powerful Qwen model. Top scores on 6 programming benchmarks. Enhanced agent coding, world knowledge, instruction following vs Qwen3.6-Plus. Free preview on Bailian. Context: 256K tokens. https://cntechpost.com/2026/04/20/alibaba-releases-qwen3-6-max-preview-stronger-instruction-following-capabilities/ ds-tracker-qw36-max-2026-04-20 Mon, 20 Apr 2026 00:00:00 +0000 Reasoning Closed Alibaba/Qwen First domain-specific OpenAI model. Life sciences: genomics, protein engineering, drug discovery. Research preview for qualified US enterprise customers. Partners: Amgen, Moderna, Allen Institute, Thermo Fisher. https://openai.com/index/introducing-gpt-rosalind/ ds-tracker-gpt-ros-2026-04-16 Thu, 16 Apr 2026 00:00:00 +0000 Reasoning Closed OpenAI #1 SWE-bench Verified (87.6%). #1 SWE-bench Pro (64.3%). 94.2% GPQA Diamond. High-res vision (3.75MP). New xhigh effort level. Agentic task budgets. $5/$25 per 1M tokens. Context: 1000K tokens. Pricing: $5/M input, $25/M output. https://www.anthropic.com/news/claude-opus-4-7 ds-tracker-co47-2026-04-16 Thu, 16 Apr 2026 00:00:00 +0000 Reasoning Closed Anthropic MoE - 3B active / 35B total, 256 experts. Runs on a laptop. 73.4% SWE-bench Verified. Natively multimodal. Extensible to 1M context. Apache 2.0. Context: 262K tokens. Open weights. https://qwen.ai/blog?id=qwen3.6-35b-a3b ds-tracker-qw36-35b-2026-04-16 Thu, 16 Apr 2026 00:00:00 +0000 Reasoning Open Alibaba/Qwen 1T MoE / 32B active. Four variants: Instant, Thinking, Agent, Agent Swarm (300 sub-agents, 4K coordinated steps). Open-weight, Modified MIT. 262K context. Context: 262K tokens. Pricing: $0.6/M input, $2.5/M output. Open weights. https://www.kimi.com/code ds-tracker-kk26-2026-04-13 Mon, 13 Apr 2026 00:00:00 +0000 Reasoning Open Moonshot AI First model from Meta Superintelligence Labs. 89.5% GPQA Diamond, 50.2% HLE (Contemplating mode). Natively multimodal with visual chain of thought and multi-agent orchestration. 262K context. Free at meta.ai. #1 on HealthBench Hard (42.8) and CharXiv Reasoning (86.4). Context: 262K tokens. https://ai.meta.com/blog/introducing-muse-spark-msl/ ds-tracker-muse-spark-2026-04-08 Wed, 08 Apr 2026 00:00:00 +0000 Reasoning Closed Meta 754B MoE / 40B active. #1 SWE-Bench Pro (58.4%), beating GPT-5.4 (57.7%) and Opus 4.6 (57.3%). 8-hour autonomous coding loops. MIT license. Context: 200K tokens. Pricing: $1.4/M input, $4.4/M output. Open weights. https://docs.z.ai/release-notes/new-released ds-tracker-glm51-2026-04-07 Tue, 07 Apr 2026 00:00:00 +0000 Reasoning Open Zhipu AI Proprietary flagship. 1M native context, 65K output tokens. 78.8% SWE-bench Verified. Always-on chain-of-thought. Native function calling and tool use. Context: 1000K tokens. Pricing: $0.29/M input, $1.65/M output. https://qwen.ai/blog?id=qwen3.6 ds-tracker-qw36-2026-04-02 Thu, 02 Apr 2026 00:00:00 +0000 Reasoning Closed Alibaba/Qwen MoE - 3.8B active / 25.2B total. Near 31B performance at a fraction of the compute. Apache 2.0. Context: 256K tokens. Open weights. https://ai.google.dev/gemma/docs/core/model_card_4 ds-tracker-ge4-26-2026-04-02 Thu, 02 Apr 2026 00:00:00 +0000 General Open Google Strongest open Gemma model. 85.2% MMLU-Pro, 84.3% GPQA Diamond. Hybrid attention with sliding window. Apache 2.0. Context: 256K tokens. Open weights. https://ai.google.dev/gemma/docs/core/model_card_4 ds-tracker-ge4-31-2026-04-02 Thu, 02 Apr 2026 00:00:00 +0000 General Open Google MiniMax March 2026 flagship. Self-evolving agentic model. 56.2% SWE-Pro. 205K context with automatic caching. $0.30/$1.20 per 1M tokens. Context: 205K tokens. Pricing: $0.3/M input, $1.2/M output. https://openrouter.ai/minimax/minimax-m2.7 ds-tracker-mm27-2026-03-18 Wed, 18 Mar 2026 00:00:00 +0000 Reasoning Closed MiniMax