Skip to main content
Entries describe the release at the time. Use the current guides for supported models, commands, and limits.
Flash 1.4.22TrainingServing
2026-11-06
  • Six training models: Added pretrained Qwen/Qwen3.5-9B-Base, with its own adapter identity. Both 9B models now use H100 for hosted FP8 serving with a 64-sequence cap. See Models.
  • Spot training: SFT/GRPO can opt into Spot with automatic recovery checkpoints. See Spot training.
  • Custom warm-start: Initialize a fresh custom optimizer from an existing adapter, or resume custom optimizer/algorithm state from an exact step. See Custom algorithms.
  • Native tool RL: The two 9B models support parsed function calls and tool results in multi-turn GRPO, including thinking. See native tool calling.
  • Private artifacts: Self-hosted controllers use Azure Blob storage for Flash-owned artifacts; Hugging Face remains optional for upstream access and explicit exports. See Self-hosting.
  • Reliability: Improved serving rollout continuity, authorization recovery, custom-session attachment, and overloaded API request cleanup.
Flash 1.4.6TrainingServing
2026-11-01
  • Custom training: All five catalog models now support text-only custom algorithms on one automatically selected Modal GPU. Use explicit target_tokens for soft-target distillation and losses over multiple tokens per position.
  • SFT tools: Include per-row OpenAI-style tools so training prompts carry the same function declarations as serving requests. Image-bearing rows do not support this field.
  • Longer context: Qwen serving now supports 262,144 tokens. Custom training remains limited to 4,096 tokens; SFT and GRPO also need a GPU shape that fits. See Context limits.
  • Hosted GPUs: The catalog uses L4, L40S, H100, and H200 serving tiers. See the model table for each model’s tier and the current token prices.
  • Token probabilities: Request generated-token scores and up to 20 alternatives with flash models chat --top-logprobs K or the Python client. See Top-K log probabilities.
  • SFT reliability: Multi-image datasets avoid the previous pixel-column overflow before training. Missing outputs and targets lost to formatting or truncation now have distinct diagnostics.
  • Recovery: Improved checkpoint-deployment recovery and run submission handling across control-plane restarts.
Flash 1.4.1TrainingServing
2026-10-30
  • Custom algorithms: Write a local Python loop with async or blocking GPU calls, custom losses, and durable resume. Start with flash custom init. See Custom algorithms.
  • Small models: Qwen3.5 0.8B and 4B support text-only SFT, GRPO, and hosted serving. See the model catalog and prices.
  • Adapter downloads: Download exact checkpoints directly with your API key, without a Hugging Face token. See Download an adapter.
  • GRPO rewards: Read rollout token probabilities and explicitly control truncation masking.
  • Images and transcripts: Image-capable models allow up to 32 images across a conversation, including multi-turn observations. SFT can use stored prompt_messages.
  • Quotes: Every algorithm now uses an authenticated control-plane quote. SFT reports actual dataset counts separately; packaged dataset profiling no longer imposes a separate 32 MiB cap.
  • Environments: Managed packages use immutable published revisions; flash env push, pull, and managed environment IDs remain the user workflow.
Flash 1.3.25ServingTraining
2026-10-24
  • DeepSeek: Use deepseek-ai/DeepSeek-V4.1-Flash-1017 for text-only chat, with optional thinking and a 32K context. See Supported models.
  • Rate limits: Hosted inference defaults to 600 requests per minute per organization. See Rate limits and retries.
  • Progress: Run status shows the current stage and elapsed timings. See Reading progress.
  • Reliability: Improved training startup, checkpoint saving, and serving.
Flash 1.3.9Serving
2026-10-21
  • Hosted serving supports thinking-enabled checkpoints and base-model requests.
  • Reasoning tokens count toward completion usage at the normal output rate.
  • Read completion_tokens_details.reasoning_tokens for the reasoning count.
See Reasoning responses.
Flash 1.3.8ServingTraining
2026-10-21
  • Use RUN_ID/final or RUN_ID/step-N for deployment, chat, evaluation, export, and warm-start. Bare run IDs are not accepted.
  • Undeploying a checkpoint leaves other checkpoints available.
  • flash models deploy --wait defaults to 4200 seconds.
  • The training catalog includes Qwen3.5 9B, Qwen3.8 27B, and Qwen3.6 35B-A3B.
  • Training and customer-owned serving use Modal.
See Deploy & chat, Supported models, and Image input limits.
Flash 1.2.104ServingTrainingCLI
2026-08-23
  • Added customer-owned serving commands: flash serve deploy, status, and undeploy.
  • Warm-start supports every pairing of SFT and GRPO.
  • Logs identify separate attempts in retried runs.
  • GRPO accepts group_size values of 2, 4, or 8, with at most 512 completions per step.
  • SFT estimates identify the published environment used for the quote.
  • Checkpoint failures and out-of-memory errors include clearer diagnostics.
  • Unknown CLI flags suggest the closest supported flag.
See Training and CLI reference.
Flash 1.2.63TrainingServingSelf-hosting
2026-08-17
  • SFT cost estimates return in the same command without starting paid training.
  • Added explicit image limits and algorithm-specific image support.
  • Added run-scoped chat and JSON/JSONL run status output.
  • flash env eval supports split and parameter overrides.
  • Removed model_revision from training configs.
See Cost and billing and Datasets.
TrainingEnvironmentsFlash CLI
2026-08-12
  • Runs can automatically select multiple GPUs; use --gpus to set a ceiling.
  • Use prompts_per_step for GRPO; batch_size is SFT-only.
  • Environment IDs use namespace/project/name.
  • Declare scorer dependencies with [environment] pip.
  • GRPO requires max_steps or max_examples for a quote.
  • Added flash-cli as an alternate command name.
See Configuration.
TrainingEnvironmentsSelf-hosting
2026-08-10
  • Login warns about non-local HTTP addresses.
See Self-hosting.
Flash CLITrainingSelf-hosting
2026-08-09
  • Added self-hosting support and configurable data and log settings.
  • Training configs reject settings that do not apply to the selected algorithm.
  • Added flash train --gpus N and configurable lora_alpha.
  • Training requires a supported catalog model.
SFT estimates changed again in Flash 1.2.63; use the current Cost and billing guide.
Flash CLIEnvironmentsServing
2026-07-31
  • Evaluate deployed models with flash env eval and held-out suites in evaluations.py.
  • Use flash models deploy --wait in scripts; it exits nonzero on failure.
  • Environment testing supports split and parameter overrides.
  • Dry-run checks your balance against the quote.
See CLI reference.
PlatformTracingServing
2026-07-28
  • Record chat completions and export them as training data with flash traces export.
  • API responses return reasoning in reasoning_content and answers in content.
See Tracing and Deploy & chat.
Flash CLIPlatformProjectsServing
2026-07-27
  • Training configs and environment commands require a project.
  • Run commands are grouped under flash runs; serving commands under flash models.
  • Manage deployments and switch projects in the dashboard.
  • Create and list projects with flash projects create and flash projects list.
PlatformProjects
2026-07-24
Organize runs, environments, and traces into projects in the platform.
PlatformTraining
2026-07-22
View training metrics, logs, and sample completions in the platform.
ServingTraining
2026-07-21
  • Improved generation and SFT training performance.
  • Multi-turn GRPO supports per-turn or per-episode scoring.
Models
2026-07-20
Added image inputs for supported training and chat tasks.
Models
2026-07-19
Added Qwen3.6-27B at the time of this release. See the current model catalog for today’s supported models.
ServingTraining
2026-07-17
Added context windows up to 32K tokens for supported training and serving models.