AI MODEL RELEASE TRACKER — LAST UPDATED AUGUST 6, 2026
● Most recent release: Qwen3.8-Max preview (August 2, 2026 via LLM Gateway)
● Most recent official release: DeepSeek V4 Flash 0731 (July 31, 2026)
● Next confirmed event: OpenAI public S-1 prospectus — mid-to-late August 2026
● Expected August: DeepSeek V4-Pro official release · Qwen3.8-Max open weights (no date) · OpenAI Codex updates
● Expected September: OpenAI IPO listing · Llama 4 Scout follow-up variants (unconfirmed)
● Free open-weight models released in 2026: Kimi K3 (Modified MIT) · DeepSeek V4 Flash (MIT) · Laguna S 2.1 (OpenMDW) · Gemma 4 (Google)
● New models arriving every: ~2 days on average across all labs per LLM Stats
Already Released in 2026 — Major Models
| Model | Lab | Date | Price input/M | Open weights |
|---|
| DeepSeek V4 Flash 0731 | DeepSeek | Jul 31 | $0.14 | ✓ MIT |
| Kimi K3 | Moonshot AI | Jul 16 (weights Jul 27) | $3.00 | ✓ Modified MIT |
| Claude Opus 5 | Anthropic | Jul 24 | $5.00 | ✗ Closed |
| Gemini 3.6 Flash | Google | Jul 21 | $1.50 | ✗ Closed |
| GPT-5.6 Luna / Terra / Sol | OpenAI | Jul 9 | $0.20 / $2.50 / $5.00 | ✗ Closed |
| Grok 4.5 | xAI | Jul 8 | $2.00 | ✗ Closed |
| Laguna S 2.1 | Poolside | Jul 2 | $0.10 | ✓ OpenMDW |
| Claude Fable 5 / Sonnet 5 | Anthropic | Jun 9 / Jun 30 | $10 / $2 | ✗ Closed |
| Gemma 4 | Google | Jun 2026 | Free | ✓ Apache 2.0 |
| Claude Opus 4.8 | Anthropic | May 2026 | $5.00 | ✗ Closed |
| Llama 4 Scout / Maverick | Meta | Apr 2026 | Free | ✓ Llama 4 License |
In Preview / Not Yet Generally Available
| Model | Lab | Status | What's missing |
|---|
| Qwen3.8-Max | Alibaba | Preview (Jul 19) | Benchmark table · per-token price · open weights · GA date |
| DeepSeek V4-Pro | DeepSeek | Preview (1.6T MoE) | Official GA date — "coming soon" per changelog |
| Claude Mythos 5 | Anthropic | Limited (approved orgs only) | Public access — suspended pending safety review per July disclosure |
| GPT-5.6 Sol Codex | OpenAI | Expanding access | General API availability — waitlist rolling out |
Expected August–September 2026
DeepSeek V4-Pro — official GA
1.6T parameter flagship. Confirmed "coming soon" in DeepSeek's July 31 changelog. No date published. Preview available via API now. V4 Flash 0731 already outperforms V4-Pro-Preview on agent benchmarks — the GA Pro will need to justify its higher price tier.
Qwen3.8-Max — open weights and GA
Alibaba promised open weights "soon" at the July 19 WAIC announcement. No Hugging Face repository exists as of August 6. No GA date, no confirmed per-token pricing, no license published. Kimi K3 shipped weights on day 11. Qwen3.8 is on day 18 with no update.
OpenAI IPO — public S-1 and listing
Not a model release but the defining capital event of August 2026. Confidential S-1 filed May 22. Public prospectus expected mid-to-late August. September listing target. Goldman Sachs, Morgan Stanley, JPMorgan leading. For information only, not investment advice.
Google Gemini 3.6 Pro / Ultra — unconfirmed
No announcement from Google. Gemini 3.6 Flash launched July 21 — Pro and Ultra variants in the same generation would follow the same pattern as Gemini 3.5 (Flash → Pro → Ultra). No timeline published.
Meta Llama 4 follow-up variants — unconfirmed
Llama 4 Scout and Maverick launched April 2026. No announcement on follow-up variants. Meta's pattern with Llama 3 suggests additional model sizes (70B, 405B equivalents) and fine-tuned variants would follow the initial launch by 3-6 months — putting potential releases in Q3-Q4 2026.
Free and Open-Weight Models in 2026
2026 has already seen more high-capability open-weight releases than any previous year. According to AI Release Tracker, the major free and open-weight releases of 2026 include:
DeepSeek V4 Flash 0731 (MIT, July 31): $0.14/M API or self-hosted. Terminal-Bench 82.7% — the cheapest capable agentic model with published benchmarks. Weights on HuggingFace at deepseek-ai/DeepSeek-V4-Flash-0731. Full review →
Kimi K3 (Modified MIT, July 27): 2.8T/104B MoE. SWE Marathon #1. BenchLM #5 of 214. Requires 8× H100 minimum to self-host. US Treasury sanctions warning on Moonshot AI — verify current status before enterprise use. Download guide →
Laguna S 2.1 (OpenMDW, July 2): 118B/8B MoE. $0.10/M on OpenRouter or self-hosted on DGX Spark (desktop). Terminal-Bench 70.2% — highest of any disclosed-size open model. Western company, no regulatory concerns. Full review →
Gemma 4 (Apache 2.0, June 2026): Google's open-weight family. Runs on everyday devices. No API cost — free to download and run locally. Full review at Gemma 4 launch →
Llama 4 Scout / Maverick (Meta, April 2026): First MoE Llamas. Natively multimodal. 10M token context on Scout. Free under Llama 4 License. Available on Hugging Face, Ollama, and most inference providers.
Qwen3.8-Max open weights (expected): Alibaba promised open weights for the 2.4T model "soon" at WAIC on July 19. No date, no license, no HuggingFace repository as of August 6, 2026.
API Deprecation Deadlines — August and Beyond
August 31, 2026: kimi-k2.5 and moonshot-v1 series (8k/32k/128k/auto) — migrate to kimi-k3. Migration guide →
August 31, 2026: Claude Sonnet 5 intro pricing ends — $2/M → $3/M input (+50%). New tokenizer adds 10-35% more tokens. Action plan →
October 24, 2026: deepseek-chat and deepseek-reasoner deprecated — migrate to deepseek-v4-flash. V4 Flash review →
How We Track AI Model Releases
This tracker is updated manually each day as part of our daily AI news coverage. We source from official lab announcements, API changelogs, and verified third-party trackers including LLM Gateway, AI Release Tracker, and Evertune. We do not add models to the "released" table until pricing, model string, and availability are confirmed from the official source. Preview or rumoured models go in the "In Preview" section with what is still missing clearly noted.
Sources: AI Release Tracker · LLM Gateway timeline · Evertune model tracker · LLM Stats · FelloAI best models August 2026 · Related: Today's AI news →