Qwen 27B Fine-tuning Context Extension
Extended fine-tuning context for Qwen 27B models to 131,072 tokens for SFT and 65,536 for DPO, with updated batch size limits.
Together AI changelog: September 2026
26 releases from to
Everything Together AI shipped in September 2026, collected from official Together AI sources. Each entry is dated and links to the page it came from.
September 2026 at a glance
Together AI introduced new serverless models with extended context lengths, launched preemptible GPU compute for cost-effective scaling, and updated CLI tools with rollouts and HIPAA compliance features. They also reduced pricing for H100 dedicated endpoints and Qwen models, while deprecating several serverless models.
Summary generated from the releases listed below.
September 2026 was Together AI's busiest month of the 12 months to then: 26 releases, about 5.8× its monthly average of 5. That was 14 releases more than August 2026. Most were Infrastructure (14), followed by Model (5) and Developer Tools (3).
Releases
26
+14 vs August 2026
Rank of 12 months
1st
Average 4.5 a month
Most shipped
Infrastructure
14 releases
Notable Together AI launches in September 2026
- Model
together/Tev1-4B-experimental Serverless Model
New serverless model with 32,768 context length and updated pricing details.
Source: Together AI Changelog - Model
MiniMaxAI/MiniMax-H3 Serverless Model
New serverless model with 2k/sec pricing and 1,000,000 context length for inference tasks.
Source: Together AI Changelog - Model
New serverless model added
The following model is now available on serverless: deepseek-ai/DeepSeek-V4.1-Flash. Pricing: $0.30 input / $1.20 output / $0.006 cached input (per 1M tokens).
Source: Together AI Changelog
All Together AI releases in September 2026
Newest first. Each entry links to the official page it came from.
September 202626 releases
LoRA Rank Limit Improvement
Increased LoRA rank limit for fine-tuning to 128 for most models, with updated CLI/SDK versions required for optimal use.
Higher LoRA rank limit for fine-tuning
You can now train low-rank adapters (LoRA) with a rank of up to 128 for the majority of models, up from 64. The default rank for these models stays at 64, so set lora_r to use a higher one.
Batch File Retention Policy Update
Batch API files now retained for 7 days after processing, with access restrictions after expiration.
Registry Name Deployment Support
Deploy fine-tuned models using registry names instead of model_object_id, simplifying deployment workflows.
Deploy a fine-tuned model by its registry name
Deploy a fine-tuned model by its registry name using the Together CLI version 2.24.0 or later. The CLI resolves the name to the same model, so you can deploy straight from the name shown in the fine-tuning jobs dashboard.
together/Tev1-4B-experimental Serverless Model
New serverless model with 32,768 context length and updated pricing details.
Qwen Model Pricing Update
Pricing reductions for Qwen models effective September 22, 2026, with updated cost structure for input/output tokens.
Pricing update
The following models have lower pricing, effective September 22, 2026. All usage from that date forward is billed at the new rates (per 1M tokens): Qwen/Qwen3.7-Max, Qwen/Qwen3.8-Flash.
Automatic Idle Shutdown for Deployments
Automatic idle shutdown feature for dedicated deployments, allowing self-scaling based on inactivity timeouts.
Automatic idle shutdown
Dedicated deployments can auto-shutdown after inactivity. Set --inactive-timeout to enable.
Together CLI v2.34.0
Version 2.34.0 of the Together CLI and Python SDK carries the rollouts release above (the SDK surface is client.beta.endpoints.rollouts) and also adds HIPAA placement support.
Model deprecations
The following models have been deprecated and are no longer available on serverless: openai/gpt-oss-27B, google/gemma-4-31B-it, thinkingmachines/Inkling-Small, intfloat/multilingual-e5-large-instruct.
HIPAA Placement Support
HIPAA-compliant placement option for deployments, restricting them to HIPAA-attested clusters for compliance requirements.
Dedicated Model Rollouts Feature
Rollouts feature for shifting traffic between deployments without changing endpoint URLs, supporting canary, blue-green, and rolling strategies.
MiniMaxAI/MiniMax-H3 Serverless Model
New serverless model with 2k/sec pricing and 1,000,000 context length for inference tasks.
New serverless model added
The following model is now available on serverless: deepseek-ai/DeepSeek-V4.1-Flash. Pricing: $0.30 input / $1.20 output / $0.006 cached input (per 1M tokens).
deepseek-ai/DeepSeek-V4.1-Flash Serverless Model
New serverless model with 1,000,000 context length, FP8 quantization, and function calling capabilities.
Together CLI v2.33.2
Version 2.33.2 of the Together CLI improves error reporting and upload feedback: Endpoint, fine-tuning, and model commands now print the API’s error message when a request fails, rather than a generic failure notice.
Preemptible GPU Compute Preview
Preemptible compute preview for GPU clusters with discounted rates, sub-hourly billing, and 5-minute drain windows for node reclamation.
Preemptible compute for GPU clusters
Preemptible compute is now in public preview for Kubernetes GPU clusters. Alongside standard nodes, you can set a preemptible GPU target, and Together provisions toward it as spare capacity becomes available, at a flat discounted rate relative to on-demand.
zai-org/GLM-5.3 Serverless Model
New serverless model available for fine-tuning with 1,000,000 context length and FP4 quantization.
Serverless Model Deprecations
Deprecation of openai/gpt-oss-20b and other models from serverless, with recommended replacements and migration guidance.
Pricing update
H100 80GB dedicated endpoint hardware is now $3.99 per hour, down from $5.49. See Dedicated endpoint pricing.
Model deprecations
The following models are deprecated and will be removed from serverless on September 14, 2026: openai/gpt-oss-20b, google/gemma-4-31B-it, thinkingmachines/Inkling-Small, intfloat/multilingual-e5-large-instruct.
H100 Dedicated Endpoint Pricing Update
Pricing reduction for H100 80GB dedicated endpoints from $5.49 to $3.99 per hour.
Where September 2026 sits
Together AI releases per month, for September 2026 and the 11 months before it. A taller bar means more releases that month; the dated entries for each month are on its own page.
Other companies in September 2026
26 of the companies we track shipped in September 2026; Together AI ranks 2nd by number of releases.
Newsletter
Stay ahead of the month's AI releases without living in every company feed.
One monthly digest of the sharpest launches, model drops, and product moves from every company we track, delivered in one clear read.
Monthly signal
The standout launches, model drops, and platform moves worth tracking.
Zero feed fatigue
One tight read instead of bouncing between company blogs and timelines.
Built for operators
Clear summaries you can scan fast and forward to the team when needed.
Monthly dispatch
Subscribe to the newsletter
Get a clean monthly dispatch covering the launches that mattered, not every minor product breadcrumb.
Monthly digestMajor launchesFast summaries
One email each month, tuned for signal over noise.