Changes to Create Training Run Route

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_max_output_tokens_per_trajectory

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_max_output_tokens_per_turn

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_max_turns_per_trajectory

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_trajectory_timeout_seconds

the options/anyOf[subschema #1: RunOptions]/max_concurrent_trajectories/anyOf[subschema #1] request property's max was increased from 512.00 to 1024.00

removed the enum value reinforce_plus_plus of the request property options/anyOf[subschema #1: RunOptions]/algorithm/anyOf[subschema #1: TrainingAlgorithm]

removed the request property options/anyOf[subschema #1: RunOptions]/batch_collection_idle_timeout_seconds

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_collection_timeout_seconds

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_enabled

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_frequency

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_max_active_rollouts

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_max_samples

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_reasoning_effort

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_samples_per_task

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_temperature

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_top_k

removed the request property options/anyOf[subschema #1: RunOptions]/evaluation_top_p

removed the request property options/anyOf[subschema #1: RunOptions]/execution_timeout_seconds

removed the request property options/anyOf[subschema #1: RunOptions]/intelligent_density_training

removed the request property options/anyOf[subschema #1: RunOptions]/max_active_rollouts

removed the request property options/anyOf[subschema #1: RunOptions]/max_output_tokens_per_step

removed the request property options/anyOf[subschema #1: RunOptions]/max_steps_off_policy

removed the request property options/anyOf[subschema #1: RunOptions]/max_total_tokens_per_trajectory

removed the request property options/anyOf[subschema #1: RunOptions]/min_lr_ratio

removed the request property options/anyOf[subschema #1: RunOptions]/min_samples_per_group

removed the request property options/anyOf[subschema #1: RunOptions]/samples_per_instance

removed the request property options/anyOf[subschema #1: RunOptions]/train_batch_size

added the new optional request property options/anyOf[subschema #1: RunOptions]/batch_size

added the new optional request property options/anyOf[subschema #1: RunOptions]/batch_timeout_seconds

added the new optional request property options/anyOf[subschema #1: RunOptions]/density_aware_training

added the new optional request property options/anyOf[subschema #1: RunOptions]/lr_min_ratio

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_concurrent_trajectories

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_output_tokens_per_trajectory

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_output_tokens_per_turn

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_staleness

added the new optional request property options/anyOf[subschema #1: RunOptions]/min_trajectories_per_group

added the new optional request property options/anyOf[subschema #1: RunOptions]/train_on_all_fail

added the new optional request property options/anyOf[subschema #1: RunOptions]/train_on_all_pass

added the new optional request property options/anyOf[subschema #1: RunOptions]/train_on_zero_advantage

added the new optional request property options/anyOf[subschema #1: RunOptions]/trajectory_timeout_seconds

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_enabled

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_final_wait_timeout_seconds

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_interval_steps

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_max_concurrent_trajectories

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_num_tasks

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_reasoning_effort

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_temperature

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_top_k

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_top_p

added the new optional request property options/anyOf[subschema #1: RunOptions]/val_trajectories_per_task

removed the enum value qwen/qwen3-235b-a22b of the request property base_model_slug

added the new optional request property options/anyOf[subschema #1: RunOptions]/load_weights_only

added the new optional request property options/anyOf[subschema #1: RunOptions]/group_size

added the new optional request property options/anyOf[subschema #1: RunOptions]/min_samples_per_group

request property options/anyOf[subschema #1: RunOptions]/samples_per_instance deprecated

the options/anyOf[subschema #1: RunOptions]/max_total_tokens_per_trajectory/anyOf[subschema #1] request property's min was decreased from 1024.00 to 1.00

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_reasoning_effort

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_temperature

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_top_k

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_top_p

added the new optional request property options/anyOf[subschema #1: RunOptions]/temperature

added the new optional request property options/anyOf[subschema #1: RunOptions]/top_k

added the new optional request property options/anyOf[subschema #1: RunOptions]/top_p

the options/anyOf[subschema #1: RunOptions]/max_total_tokens_per_trajectory/anyOf[subschema #1] request property's max was increased from 262144.00 to 2097152.00

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_frequency

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_steps_off_policy

header request parameter Idempotency-Key list-of-types was narrowed by removing types null

for the header request parameter Idempotency-Key, the minLength was increased from 0 to 1

for the header request parameter Idempotency-Key, the maxLength was set to 256

added the new optional request property options/anyOf[subschema #1: RunOptions]/evaluation_enabled

for the header request parameter Idempotency-Key, the type was generalized from no type to string

removed the enum value none of the request property options/anyOf[subschema #1: RunOptions]/reasoning_effort/anyOf[subschema #1]

removed the request property options/anyOf[subschema #1: RunOptions]/disable_thinking

removed the request property options/anyOf[subschema #1: RunOptions]/n_parallel_agents

added the new off enum value to the request property options/anyOf[subschema #1: RunOptions]/reasoning_effort/anyOf[subschema #1]

added the new on enum value to the request property options/anyOf[subschema #1: RunOptions]/reasoning_effort/anyOf[subschema #1]

added the new optional request property options/anyOf[subschema #1: RunOptions]/max_active_rollouts

added the new optional request property options/anyOf[subschema #1: RunOptions]/reasoning_effort

added the new openai/gpt-6-astra enum value to the request property base_model_slug

added the new openai/gpt-6-sol enum value to the request property base_model_slug

added the new openai/gpt-6.1-sol enum value to the request property base_model_slug

removed the request property options/anyOf[subschema #1: RunOptions]/max_response_chars_per_tool_call

removed the request property options/anyOf[subschema #1: RunOptions]/max_tool_calls_per_step

removed the request property options/anyOf[subschema #1: RunOptions]/train_on_ungraded

added the new optional request property options/anyOf[subschema #1: RunOptions]/train_on_error

added the new optional request property options/anyOf[subschema #1: RunOptions]/train_on_timeout

added the new z-ai/glm-5.3-flash enum value to the request property base_model_slug

removed the enum value Qwen/Qwen3.6-27B of the request property base_model_slug

the options/anyOf[subschema #1: RunOptions]/lr_hold_steps/anyOf[subschema #1] request property's max was increased from 400.00 to 100000.00

the options/anyOf[subschema #1: RunOptions]/lr_warmup_steps/anyOf[subschema #1] request property's max was increased from 400.00 to 100000.00

the options/anyOf[subschema #1: RunOptions]/num_steps/anyOf[subschema #1] request property's max was increased from 400.00 to 100000.00

removed the request property options/anyOf[subschema #1: RunOptions]/retries

added the new optional request property options/anyOf[subschema #1: RunOptions]/retries

added the new optional request property agent_name

added the new optional request property benchmark_name

added the new optional request property options/anyOf[subschema #1: RunOptions]/intelligent_density_training

the request property bench_id became optional

request property bench_id list-of-types was widened by adding types null to media type application/json

added the new required request property base_model_slug

removed the request property base_model_id

removed the request property training_options

added the new optional request property bypass_ownership

added the new optional request property display_name

added the new optional request property options

the training_options/max_response_chars_per_tool_call/anyOf[subschema #1] request property's min was increased to 128.00

the training_options/max_total_tokens_per_trajectory/anyOf[subschema #1] request property's min was increased to 1024.00

the training_options/samples_per_instance/anyOf[subschema #1] request property's min was increased to 4.00

the training_options/learning_rate/anyOf[subschema #1] request property's max was set to 0.01

the training_options/lr_hold_steps/anyOf[subschema #1] request property's max was set to 400.00

the training_options/lr_warmup_steps/anyOf[subschema #1] request property's max was set to 400.00

the training_options/max_output_tokens_per_step/anyOf[subschema #1] request property's max was set to 262144.00

the training_options/max_response_chars_per_tool_call/anyOf[subschema #1] request property's max was set to 10000000.00

the training_options/max_tool_calls_per_step/anyOf[subschema #1] request property's max was set to 20.00

the training_options/max_total_tokens_per_trajectory/anyOf[subschema #1] request property's max was set to 262144.00

the training_options/max_turns_per_trajectory/anyOf[subschema #1] request property's max was set to 300.00

the training_options/n_parallel_agents/anyOf[subschema #1] request property's max was set to 512.00

the training_options/num_steps/anyOf[subschema #1] request property's max was set to 400.00

the training_options/samples_per_instance/anyOf[subschema #1] request property's max was set to 32.00

the training_options/train_batch_size/anyOf[subschema #1] request property's max was set to 256.00

the training_options/learning_rate/anyOf[subschema #1] request property's min was set to 0.00

removed the request property training_options/eval_final_collection_timeout_seconds

removed the request property training_options/overrides

added the new optional request property training_options/evaluation_collection_timeout_seconds

added the new optional request property training_options/evaluation_max_active_rollouts

added the new optional request property training_options/evaluation_max_samples

added the new optional request property training_options/evaluation_samples_per_task

added the new qwen/qwen3-235b-a22b enum value to the request property base_model_id

added the new moonshotai/kimi-k3 enum value to the request property base_model_id

added the new z-ai/glm-5.3 enum value to the request property base_model_id

added the new anthropic/claude-opus-5.5 enum value to the request property base_model_id

added the new anthropic/claude-sonnet-5.5 enum value to the request property base_model_id

added the new openai/gpt-5.6-luna enum value to the request property base_model_id

added the new openai/gpt-5.6-sol enum value to the request property base_model_id

added the new thinkingmachines/Inkling-Small enum value to the request property base_model_id

added the new optional request property agent_id

the request property training_options became optional

the request property training_options became required

the optional response header X-Trajectory-Training-Created removed for the status 202

removed the request property training_options/max_active_rollouts

removed the request property training_options/reject_all_fail

removed the request property training_options/reject_all_pass

removed the request property training_options/reject_zero_advantage

added the new optional request property training_options/max_tool_calls_per_step

added the new optional request property training_options/max_total_tokens_per_trajectory

added the new optional request property training_options/n_parallel_agents

added the new optional request property training_options/overrides

the batch_collection_idle_timeout_seconds request property default value 9600 was removed

the disable_thinking request property default value false was removed

the eval_final_collection_timeout_seconds request property default value 7200 was removed

the execution_timeout_seconds request property default value 7200 was removed

the learning_rate request property default value 0.0001 was removed

the lr_hold_steps request property default value 135 was removed

the lr_warmup_steps request property default value 15 was removed

the max_output_tokens_per_step request property default value 16384 was removed

the max_response_chars_per_tool_call request property default value 16384 was removed

the max_turns_per_trajectory request property default value 125 was removed

the min_lr_ratio request property default value 0.1 was removed

the num_steps request property default value 100 was removed

the samples_per_instance request property default value 8 was removed

the train_batch_size request property default value 32 was removed

request property training_options/batch_collection_idle_timeout_seconds list-of-types was widened by adding types null to media type application/json

request property training_options/disable_thinking list-of-types was widened by adding types null to media type application/json

request property training_options/eval_final_collection_timeout_seconds list-of-types was widened by adding types null to media type application/json

request property training_options/execution_timeout_seconds list-of-types was widened by adding types null to media type application/json

request property training_options/learning_rate list-of-types was widened by adding types null to media type application/json

request property training_options/lr_hold_steps list-of-types was widened by adding types null to media type application/json

request property training_options/lr_warmup_steps list-of-types was widened by adding types null to media type application/json

request property training_options/max_output_tokens_per_step list-of-types was widened by adding types null to media type application/json

request property training_options/max_response_chars_per_tool_call list-of-types was widened by adding types null to media type application/json

request property training_options/max_turns_per_trajectory list-of-types was widened by adding types null to media type application/json

request property training_options/min_lr_ratio list-of-types was widened by adding types null to media type application/json

request property training_options/num_steps list-of-types was widened by adding types null to media type application/json

request property training_options/samples_per_instance list-of-types was widened by adding types null to media type application/json

request property training_options/train_batch_size list-of-types was widened by adding types null to media type application/json

request property training_options/batch_collection_idle_timeout_seconds list-of-types was narrowed by removing types null from media type application/json

request property training_options/disable_thinking list-of-types was narrowed by removing types null from media type application/json

request property training_options/eval_final_collection_timeout_seconds list-of-types was narrowed by removing types null from media type application/json

request property training_options/execution_timeout_seconds list-of-types was narrowed by removing types null from media type application/json

request property training_options/learning_rate list-of-types was narrowed by removing types null from media type application/json

request property training_options/lr_hold_steps list-of-types was narrowed by removing types null from media type application/json

request property training_options/lr_warmup_steps list-of-types was narrowed by removing types null from media type application/json

request property training_options/max_output_tokens_per_step list-of-types was narrowed by removing types null from media type application/json

request property training_options/max_response_chars_per_tool_call list-of-types was narrowed by removing types null from media type application/json

request property training_options/max_turns_per_trajectory list-of-types was narrowed by removing types null from media type application/json

request property training_options/min_lr_ratio list-of-types was narrowed by removing types null from media type application/json

request property training_options/num_steps list-of-types was narrowed by removing types null from media type application/json

request property training_options/samples_per_instance list-of-types was narrowed by removing types null from media type application/json

request property training_options/train_batch_size list-of-types was narrowed by removing types null from media type application/json

the training_options/batch_collection_idle_timeout_seconds request property's max was set to 38400.00

the training_options/eval_final_collection_timeout_seconds request property's max was set to 38400.00

the training_options/execution_timeout_seconds request property's max was set to 36000.00

the training_options/learning_rate request property's max was set to 0.01

the training_options/lr_hold_steps request property's max was set to 400.00

the training_options/lr_warmup_steps request property's max was set to 400.00

the training_options/max_output_tokens_per_step request property's max was set to 262112.00

the training_options/max_response_chars_per_tool_call request property's max was set to 65536.00

the training_options/max_turns_per_trajectory request property's max was set to 300.00

the training_options/min_lr_ratio request property's max was set to 1.00

the training_options/num_steps request property's max was set to 400.00

the training_options/samples_per_instance request property's max was set to 64.00

the training_options/train_batch_size request property's max was set to 256.00

the training_options/batch_collection_idle_timeout_seconds request property's min was set to 1.00

the training_options/eval_final_collection_timeout_seconds request property's min was set to 1.00

the training_options/execution_timeout_seconds request property's min was set to 1.00

the training_options/learning_rate request property's min was set to 0.00

the training_options/lr_hold_steps request property's min was set to 0.00

the training_options/lr_warmup_steps request property's min was set to 0.00

the training_options/max_output_tokens_per_step request property's min was set to 1.00

the training_options/max_response_chars_per_tool_call request property's min was set to 128.00

the training_options/max_turns_per_trajectory request property's min was set to 1.00

the training_options/min_lr_ratio request property's min was set to 0.00

the training_options/num_steps request property's min was set to 1.00

the training_options/samples_per_instance request property's min was set to 4.00

the training_options/train_batch_size request property's min was set to 1.00

removed the request property training_options/max_tool_calls_per_step

removed the request property training_options/max_total_tokens_per_trajectory

removed the request property training_options/n_parallel_agents

removed the request property training_options/overrides

added the new optional request property training_options/max_active_rollouts

added the new optional request property training_options/reject_all_fail

added the new optional request property training_options/reject_all_pass

added the new optional request property training_options/reject_zero_advantage

the request property training_options became optional

the batch_collection_idle_timeout_seconds request property default value 9600 was added

the disable_thinking request property default value false was added

the eval_final_collection_timeout_seconds request property default value 7200 was added

the execution_timeout_seconds request property default value 7200 was added

the learning_rate request property default value 0.0001 was added

the lr_hold_steps request property default value 135 was added

the lr_warmup_steps request property default value 15 was added

the max_output_tokens_per_step request property default value 16384 was added

the max_response_chars_per_tool_call request property default value 16384 was added

the max_turns_per_trajectory request property default value 125 was added

the min_lr_ratio request property default value 0.1 was added

the num_steps request property default value 100 was added

the samples_per_instance request property default value 8 was added

the train_batch_size request property default value 32 was added

added the new optional request property training_options/overrides

removed the request property training_options/reject_all_fail

removed the request property training_options/reject_all_pass

removed the request property training_options/reject_zero_advantage

removed the request property training_options/samples_per_prompt

added the new optional request property training_options/samples_per_instance

added the new Qwen/Qwen3.6-27B enum value to the request property base_model_id

added the new Qwen/Qwen3.8-27B enum value to the request property base_model_id

added the new optional request property training_options/reject_all_fail

added the new optional request property training_options/reject_all_pass

added the new optional request property training_options/reject_zero_advantage

added the new optional request property training_options/samples_per_prompt

the training_options/eval_final_collection_timeout_seconds/anyOf[subschema #1] request property's max was decreased to 38400.00

the training_options/execution_timeout_seconds/anyOf[subschema #1] request property's max was decreased to 36000.00

removed the required property execution_id from the response with the 202 status

removed the required property experiment_name from the response with the 202 status

removed the required property pipeline_name from the response with the 202 status

removed the required property pipeline_run_id from the response with the 202 status

removed the optional property resolved_training_options from the response with the 202 status

added the new optional request property parent_checkpoint_id

the training_options/eval_final_collection_timeout_seconds/anyOf[subschema #1] request property's max was increased from 38400.00 to 96000.00

the training_options/execution_timeout_seconds/anyOf[subschema #1] request property's max was increased from 36000.00 to 93600.00

added the optional property resolved_training_options to the response with the 202 status

removed the optional property resolved_training_options from the response with the 202 status

removed the request property training_options/no_progress_timeout_seconds

added the new optional request property training_options/batch_collection_idle_timeout_seconds

the disable_thinking request property default value false was removed

request property training_options/disable_thinking list-of-types was widened by adding types null to media type application/json

the training_options/eval_final_collection_timeout_seconds/anyOf[subschema #1] request property's max was increased from 36000.00 to 38400.00

the training_options/no_progress_timeout_seconds/anyOf[subschema #1] request property's max was increased from 36000.00 to 38400.00

added the optional property resolved_training_options to the response with the 202 status

added the new optional request property training_options/eval_final_collection_timeout_seconds

added the new optional request property training_options/execution_timeout_seconds

added the new optional request property training_options/learning_rate

added the new optional request property training_options/lr_hold_steps

added the new optional request property training_options/lr_warmup_steps

added the new optional request property training_options/min_lr_ratio

added the new optional request property training_options/no_progress_timeout_seconds

response property execution_id list-of-types was widened by adding types null to media type application/json of response 202

response property pipeline_name list-of-types was widened by adding types null to media type application/json of response 202

response property pipeline_run_id list-of-types was widened by adding types null to media type application/json of response 202

the resolved_trajectory_options response's property type changed from object to no type for status 202

removed the required property resolved_trajectory_options/max_output_tokens_per_step from the response with the 202 status

removed the required property resolved_trajectory_options/max_response_chars_per_tool_call from the response with the 202 status

removed the required property resolved_trajectory_options/max_tool_calls_per_step from the response with the 202 status

removed the required property resolved_trajectory_options/max_total_tokens_per_trajectory from the response with the 202 status

removed the required property resolved_trajectory_options/max_turns_per_trajectory from the response with the 202 status

removed the required property resolved_trajectory_options/timeout_seconds from the response with the 202 status

added the new optional header request parameter Idempotency-Key

added ResolvedTrajectoryTrainingOptions subschema #2 to the resolved_trajectory_options response property anyOf list for the response status 202

the request property training_options became required

added the new optional request property training_options/disable_thinking

added the new optional request property training_options/n_parallel_agents

added the new optional request property training_options/train_batch_size

added the new optional request property training_options/num_steps