Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Original file line number Diff line number Diff line change
@@ -0,0 +1,58 @@
# Sources (checked 2026-11-16):
# - Public aliases, admission budget, output cap, tools, effort and request fields:
# https://oathbreaker.ai/docs
# - USD per million tokens: https://oathbreaker.ai/pricing
# - Public availability and current limits/prices: https://oathbreaker.ai/llms.txt
# Oathbreaker serves an abliterated variant of DeepSeek V4.1 Flash 1017 through its
# own gateway; it does not train the underlying DeepSeek model.
# The exact served derivative is not published as downloadable Oathbreaker weights.
# Lab weights and benchmarks are never inherited by provider entries; the canonical
# model link identifies the DeepSeek V4.1 Flash 1017 lineage, not equivalent results.
# release_date is when this alias first appeared in Oathbreaker's public llms.txt
# (observed 2026-11-16), not the base model launch.
# Limits: a request is admitted when UTF-8 input bytes, message and tool framing and
# the requested output tokens fit in a 131,072-unit budget, and the request body is at
# most 262,144 bytes (limit.request_bytes). The budget counts bytes, not tokens, so
# limit.context is a conservative token estimate: 131,072 / ~4 bytes per token of typical
# English text = 32,768. Code or non-English text can use more bytes per token.
# Effort: reasoning_effort = low|high|max. If the field is omitted, the gateway sends no
# effort parameter and the upstream model default applies; no default is documented.
# "none" is not an accepted value, and no thinking toggle field is accepted.
# max_tokens and max_completion_tokens are supported (default 8,192); output includes reasoning.
# Cache reads are implicit and billed only when reported by upstream usage.
# Chat Completions only; no image input, Responses, Messages or strict tool schemas.
# This alias serves the same model, limits and prices as oathbreaker-deepseek-v4.1-flash
# on a separate serving route.
schema = 2
base_model = "deepseek/deepseek-v4.1-flash-1017"
name = "Oathbreaker DeepSeek V4.1 Flash Abliterated Fast"
description = "DeepSeek V4.1 Flash 1017 Abliterated on a separate Oathbreaker serving route, through its text-only OpenAI-compatible Chat Completions API"
open_weights = false
release_date = "2026-11-16"

[modalities]
input = ["text"]
output = ["text"]

[capabilities]
tools = { supported = true, choice = ["auto", "none", "required", "tool"] }
reasoning = { supported = true, effort = ["low", "high", "max"] }
structured_output = false
top_k = false
cache = { implicit = true }
stream_usage = true

[limit]
context = 32_768
output = 32_768
request_bytes = 262_144

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 0.50
cache = { read = 0.05 }

[cost.text.output]
price = 1.50
56 changes: 56 additions & 0 deletions providers/oathbreaker/models/oathbreaker-deepseek-v4.1-flash.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,56 @@
# Sources (checked 2026-11-16):
# - Public aliases, admission budget, output cap, tools, effort and request fields:
# https://oathbreaker.ai/docs
# - USD per million tokens: https://oathbreaker.ai/pricing
# - Public availability and current limits/prices: https://oathbreaker.ai/llms.txt
# Oathbreaker serves an abliterated variant of DeepSeek V4.1 Flash 1017 through its
# own gateway; it does not train the underlying DeepSeek model.
# The exact served derivative is not published as downloadable Oathbreaker weights.
# Lab weights and benchmarks are never inherited by provider entries; the canonical
# model link identifies the DeepSeek V4.1 Flash 1017 lineage, not equivalent results.
# release_date is when this alias first appeared in Oathbreaker's public llms.txt
# (observed 2026-11-16), not the base model launch.
# Limits: a request is admitted when UTF-8 input bytes, message and tool framing and
# the requested output tokens fit in a 131,072-unit budget, and the request body is at
# most 262,144 bytes (limit.request_bytes). The budget counts bytes, not tokens, so
# limit.context is a conservative token estimate: 131,072 / ~4 bytes per token of typical
# English text = 32,768. Code or non-English text can use more bytes per token.
# Effort: reasoning_effort = low|high|max. If the field is omitted, the gateway sends no
# effort parameter and the upstream model default applies; no default is documented.
# "none" is not an accepted value, and no thinking toggle field is accepted.
# max_tokens and max_completion_tokens are supported (default 8,192); output includes reasoning.
# Cache reads are implicit and billed only when reported by upstream usage.
# Chat Completions only; no image input, Responses, Messages or strict tool schemas.
schema = 2
base_model = "deepseek/deepseek-v4.1-flash-1017"
name = "Oathbreaker DeepSeek V4.1 Flash Abliterated"
description = "DeepSeek V4.1 Flash 1017 Abliterated served through Oathbreaker's text-only OpenAI-compatible Chat Completions API"
open_weights = false
release_date = "2026-11-16"

[modalities]
input = ["text"]
output = ["text"]

[capabilities]
tools = { supported = true, choice = ["auto", "none", "required", "tool"] }
reasoning = { supported = true, effort = ["low", "high", "max"] }
structured_output = false
top_k = false
cache = { implicit = true }
stream_usage = true

[limit]
context = 32_768
output = 32_768
request_bytes = 262_144

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 0.50
cache = { read = 0.05 }

[cost.text.output]
price = 1.50
Loading