What has a profile in the last 28 days

Window ends August 15, 2026. For announcements, previews, availability changes, and releases still awaiting a full profile, see the full release timeline. Need the comparison table too? Compare models · 2-min chooser · Stack cost calculator · Popular this week · Source receipts.

FilterAllZ.aiQwenGoogleSpaceXAIMetaDeepSeekAnthropicInclusionAI

TypeOpen weightsfrontierbalancedbudgetModalityaudioimagepdftextvideo

5 releases4 providers5 open-weight

Z.aiQwenGoogleSpaceXAIMetaDeepSeekAnthropicInclusionAI

1. 

   Qwen3.8-27B

   BalancedOpen weights

   Qwen

   Released yesterdayDate released: August 14, 2026

   Apache 2.0 Qwen3.8 dense VLM for local and self-hosted work — native 262K context, image and video input, thinking on by default but can be turned off. Distinct from hosted Qwen3.8-Max.

   • text
   • image
   • video

   Best for: Local multimodal agents · Self-hosting a dense 27B VLM · Apache 2.0 deployments

   Watch out: The Hugging Face repo was created 2026-08-05; this row uses the 2026-08-14 card revision. Hosted 1M-context API is documented as coming soon. Vendor SWE-bench Pro 61.7 and DeepSWE 42.2 used a Claude Code harness, not swebench.com or the public Datacurve DeepSWE board. YaRN can extend context toward 1M.

   View in model comparison tableRun the 2-minute chooser

   Released

   August 14, 2026

   Input / 1M

   Not verified

   Output / 1M

   Not verified

   Context

   262K

   Max output

   131K

2. 

   Qwen3.8-2.4T-A95B

   FrontierOpen weights

   Qwen

   Released 3 days agoDate released: August 12, 2026

   Official Hugging Face checkpoint for Qwen3.8 open weights — a text base model, not the hosted Qwen3.8-Max API.

   • text

   Best for: Self-hosting the Qwen3.8 flagship · vLLM / SGLang serving · Evaluating the open Max-class weights

   Watch out: Native context is 262,144 tokens (YaRN extends toward ~1,010,000). Hugging Face best-practice caps: 262,144 reasoning tokens and 131,072 final-response tokens. Thinking cannot be disabled. Qwen3.8-Max License: MaaS / AI Work Assistant businesses over $50M trailing 12-month revenue need a separate commercial licence. Self-hosting still needs datacentre-class memory. Do not treat this checkpoint as the multimodal 1M-context API.

   View in model comparison tableRun the 2-minute chooser

   Released

   August 12, 2026

   Input / 1M

   Not verified

   Output / 1M

   Not verified

   Context

   262K

   Max output

   131K

3. 

   Muse Glimmer 30B

   BalancedOpen weights

   Meta

   Released 5 days agoDate released: August 10, 2026

   Meta's open-weight multimodal agentic model distilled from Muse Spark for local consumer hardware (~24–32GB class with 4-bit).

   • text
   • image

   Best for: Local agents · On-device coding + tool use · Privacy-sensitive multimodal work

   Watch out: Full BF16 needs far more than 24GB — plan on official GGUF / 4-bit packs (and mmproj for vision). Agentic quality ≠ frontier Muse Spark API; validate on your workflows.

   View in model comparison tableRun the 2-minute chooser

   Released

   August 10, 2026

   Input / 1M

   $0

   Output / 1M

   $0

   Context

   131K

   Max output

   Not verified

4. 

   DeepSeek V4-Flash

   BudgetOpen weights

   DeepSeek

   Released 15 days agoDate released: July 31, 2026

   Open-weight price-performance pick: MIT weights, 1M context, and API rates far below closed budget tiers.

   • text

   Best for: Cost-sensitive hosted agents · High-volume coding assist · Open-weight deployments with cluster VRAM

   Watch out: API list prices move to peak/off-peak from 2026-08-16 16:00 UTC (see pricing note). Self-hosting the ~284B MoE still needs roughly 90–170GB+ class VRAM depending on quant — not a laptop budget build. DeepSWE snapshot also uses the shared mini-swe-agent harness — verify cost, effort, and serving conditions before treating the result as a forecast.

   View in model comparison tableRun the 2-minute chooser

   Released

   July 31, 2026

   Input / 1M

   $0.14

   Output / 1M

   $0.28

   Context

   1M

   Max output

   384K

5. 

   Ling 3.0 Flash

   BudgetOpen weights

   InclusionAI

   Released 22 days agoDate released: July 24, 2026

   InclusionAI's cost-focused Flash tier for high-frequency hybrid reasoning and agent loops at low list rates.

   • text

   Best for: High-volume agent loops · Cost-sensitive chat and extraction · Fast hybrid reasoning

   Watch out: Native context is 256Ki tokens (262,144). Confirm serving provider and current list before committing volume.

   View in model comparison tableRun the 2-minute chooser

   Released

   July 24, 2026

   Input / 1M

   $0.075

   Output / 1M

   $0.22

   Context

   262K

   Max output

   Not verified

vs

GPT-5.6 Sol vs Qwen3.8-2.4T-A95B

vs

GPT-5.6 Terra vs Qwen3.8-27B

vs

GPT-5.6 Terra vs Muse Glimmer 30B

vs

DeepSeek V4-Flash vs GPT-5.6 Luna

vs

GPT-5.6 Luna vs Ling 3.0 Flash

vs

Claude Opus 5 vs Qwen3.8-2.4T-A95B

vs

Claude Fable 5 vs Qwen3.8-2.4T-A95B

vs

Claude Sonnet 5 vs Qwen3.8-27B

vs

Claude Sonnet 5 vs Muse Glimmer 30B

vs

Gemini 3.1 Pro vs Qwen3.8-2.4T-A95B

vs

DeepSeek V4-Flash vs Gemini 3.6 Flash

vs

Gemini 3.6 Flash vs Ling 3.0 Flash