Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
097ddd2
Add HopBase provider
victor-zzh Oct 31, 2026
019388f
hopbase: address review (reasoning evidence, freeform, billing notes)
victor-zzh Oct 31, 2026
a1a6352
hopbase: add the >200K long-context tier to gemini-3.1-pro-preview
victor-zzh Nov 5, 2026
3374314
hopbase: add claude-fable-5-2 (Anthropic Messages)
victor-zzh Nov 10, 2026
80aa560
Merge branch 'dev' into add-hopbase
victor-zzh Nov 11, 2026
727c769
fix(hopbase): use the generic anthropic-compatible opencodeai entry f…
victor-zzh Nov 11, 2026
cdb1e7b
docs(hopbase): clarify that only explicit caching is declared for Cla…
victor-zzh Nov 11, 2026
32de995
hopbase: claude-fable-5-2 declares explicit cache only
victor-zzh Nov 13, 2026
c1c7dc5
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
0af9926
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
02331b7
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
367819d
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
4ebe69b
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
ec9698c
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
b281e46
chore(hopbase): simplify doubao pricing comments
victor-zzh Nov 14, 2026
2be785f
hopbase: declare implicit caching for gemini-3.8-flash
victor-zzh Nov 15, 2026
424845b
hopbase: declare implicit caching for gemini-3.1-pro-preview
victor-zzh Nov 15, 2026
6d89f60
hopbase: declare implicit caching for gemini-3.1-pro-preview
victor-zzh Nov 15, 2026
d88f899
hopbase: add kimi-k3.1-pro (verified live 2026-11-17)
victor-zzh Nov 16, 2026
74f632a
Merge remote-tracking branch 'upstream/dev' into add-hopbase
victor-zzh Nov 16, 2026
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
45 changes: 45 additions & 0 deletions models/bytedance-seed/seed-2.1-pro-0915.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,45 @@
# Sources (accessed 2026-10-31):
# - Snapshot ID doubao-seed-2-1-pro-260915 (YYMMDD suffix = 2026-09-15), the newer Seed 2.1 Pro
# snapshot on Volcengine Ark: https://www.volcengine.com/docs/82379/1330310
# - Limits: 1M context, up to 256K output; the earlier 260628 snapshot is 256K context
# (bytedance-seed/seed-2.1-pro).
# - Image/video formats: same Ark image and video understanding tables as seed-2.1-pro:
# https://www.volcengine.com/docs/82379/1362931, https://www.volcengine.com/docs/82379/1895586
# - Sampling: the Ark Chat API takes temperature and top_p and has no top_k: https://www.volcengine.com/docs/82379/1494384
schema = 2
name = "Seed 2.1 Pro 0915"
description = "September 2026 snapshot of ByteDance's flagship Seed 2.1 Pro with a 1M context window for multimodal reasoning, coding, and agents"
family = "seed"
open_weights = false
release_date = "2026-09-15"

[modalities]
input = [
"text",
"image/jpeg",
"image/png",
"image/gif",
"image/webp",
"image/bmp",
"image/tiff",
"image/x-icon",
"image/x-icns",
"image/sgi",
"image/jp2",
"image/heic",
"video/mp4",
"video/x-msvideo",
"video/quicktime",
]
output = ["text"]

[capabilities]
tools = { supported = true }
reasoning = { supported = true }
temperature = true
top_p = true
top_k = false

[limit]
context = 1_048_576
output = 262_144
28 changes: 28 additions & 0 deletions models/kuaishou/kling-v4.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,28 @@
# Released 2026-10-15: https://bernama.com/en/press/news.php?id=2622471
# ("Kling AI Officially Launches Next-Generation Kling 4.0").
# Inputs: text prompt plus first/last-frame images (JPG/JPEG or PNG, as for Kling 3.0):
# https://app.klingai.com/global/dev/document-api/apiReference/model/imageToVideo
# Output: MP4; generated sound is a track inside the MP4, so no audio MIME type is listed.
# Sampling: the task API has no temperature, top_p or top_k.
schema = 2
v2_only = true
type = "video"
name = "Kling VIDEO 4.0"
description = "Kling 4.0 video model with clips up to 30 seconds, keyframe control and native audio"
open_weights = false
release_date = "2026-10-15"

[modalities]
input = ["text", "image/jpeg", "image/png"]
output = ["video/mp4"]

[capabilities]
tools = { supported = false }
reasoning = { supported = false }
temperature = false
top_p = false
top_k = false

[limit]
context = 0
output = 0
26 changes: 26 additions & 0 deletions models/xai/grok-4.20-multi-agent-0309.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,26 @@
# Sources (docs.x.ai), mirrored from providers/xai/models/grok-4.20-multi-agent-0309.toml:
# - Model page: https://docs.x.ai/developers/models/grok-4.20-multi-agent-0309
# - Client-side function tools need beta access on xAI ("Client-side tools for multi-agent models
# require beta access"), so tools stay off for general access.
# Lab file added so relays can point base_model at it; the xAI provider file stays a full definition.
schema = 2
name = "Grok 4.20 Multi-Agent"
description = "Grok model for agentic tool use, reasoning, coding, and live assistance"
family = "grok"
open_weights = false
release_date = "2026-03-09"

[modalities]
input = ["text", "image/jpeg", "image/png", "application/pdf"]
output = ["text"]

[capabilities]
tools = { supported = false }
reasoning = { supported = true }
temperature = true
top_p = true
top_k = true

[limit]
context = 1_000_000
output = 30_000
5 changes: 5 additions & 0 deletions providers/hopbase/logo.svg
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
32 changes: 32 additions & 0 deletions providers/hopbase/models/claude-fable-5-2.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
# Verified live 2026-11-10 against POST https://api.hop-base.com/v1/messages (Anthropic Messages,
# x-api-key auth) with this exact model ID and a regular customer API key. HopBase serves Claude
# models over Messages only, so this file overrides the provider's Chat Completions endpoint.
# Streaming returned message_delta usage and message_stop. tool_choice auto returned a well-formed
# tool_use; a strict tool returned tool_use; tool_choice any returns HTTP 400 "tool_choice: type
# \"tool\" and \"any\" are not supported for this model", as on Anthropic.
# Effort: output_config.effort = low|medium|high|xhigh|max all accepted.
# Images: a red 64x64 PNG returned "Red".
# Cache: cache_control on the system block wrote 6,026 tokens and the repeat read 6,026. Only explicit
# caching is declared: HopBase documents cache_control markers and makes no promise for prompts sent
# without them.
# Pricing: price any registered account pays in the public "Claude Max (ccmax)" group, USD
# (official $8 / $40, $0.20 cache read, $10 5m cache write, https://platform.claude.com/docs/en/about-claude/pricing,
# times the group rate, rounded up). Model page: https://hop-base.com/en/docs/models/claude-fable-5-2
schema = 2
base_model = "anthropic/claude-fable-5-2"

[capabilities]
tools = { supported = true, strict = true, choice = ["auto", "none"] }
reasoning = { supported = true, effort = ["low", "medium", "high", "xhigh", "max"] }
cache = { explicit = { ttl = ["5m"], max_breakpoints = 4, targets = ["tools", "system", "messages"] } }

[api.messages]
base_url = "https://api.hop-base.com/v1"
sdk = { aisdk = "@ai-sdk/anthropic", opencodeai = "@opencode/ai/providers/anthropic-compatible" }

[cost.text.input]
price = 2.24
cache = { read = 0.056, write = 2.8 }

[cost.text.output]
price = 11.18
38 changes: 38 additions & 0 deletions providers/hopbase/models/deepseek-v4-pro-202606.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,38 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# HopBase serves DeepSeek V4 Pro 0813 under the ID deepseek-v4-pro-202606 (https://hop-base.com/llms.txt).
# Limits: 1,048,576-token context, up to 393,216 output tokens (same source). Text only: an image
# returns HTTP 400 image_input_not_supported before reaching the model.
# Toggle: thinking.type = enabled|disabled (disabled -> 0 reasoning tokens).
# Effort: reasoning_effort = low|high|max, the native DeepSeek levels; medium is accepted and
# mapped by DeepSeek, none also turns thinking off, an unknown value returns HTTP 400.
# Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Structured output: json_schema with strict = true returned an enum-valid object.
# Cache: repeating a ~2.5K-token prefix returned cached_tokens = 2048.
# Pricing: HopBase self-serve price, USD (https://hop-base.com/pricing, also in https://hop-base.com/llms.txt). Single all-day rate:
# HopBase does not apply DeepSeek's peak multiplier to this ID (billing logs: 3,037 requests over
# the 14 days to 2026-10-31, every one at the same per-token price across peak and off-peak hours).
schema = 2
base_model = "deepseek/deepseek-v4-pro-0813"

[capabilities]
reasoning = { supported = true, toggle = true, effort = ["low", "high", "max"] }
structured_output = true
stream_usage = true
cache = { implicit = true }

[limit]
context = 1_048_576
output = 393_216

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

[cost.text.input]
price = 0.726
cache = { read = 0.0242 }

[cost.text.output]
price = 2.178
46 changes: 46 additions & 0 deletions providers/hopbase/models/deepseek-v4.1-flash-1017-sg.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,46 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# Same DeepSeek V4.1 Flash 2026-10-17 release, served from a Singapore data center under its own ID.
# Limits: 1M context, up to 393,216 output tokens (https://hop-base.com/llms.txt).
# Toggle: thinking.type = enabled|disabled (disabled -> 0 reasoning tokens).
# Effort: reasoning_effort = low|high|max, the native DeepSeek levels; medium is accepted and
# mapped by DeepSeek, none also turns thinking off, an unknown value returns HTTP 400.
# Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Images: a red PNG data URL returned "Red". Structured output: json_schema with strict = true
# returned an enum-valid object. Cache: a repeated ~2.4K-token prefix returned
# prompt_tokens_details.cached_tokens = 2432. tool_choice none|required|named all honoured.
# Also checked through @ai-sdk/openai-compatible 1.0.60 (generateText, streamText, tool round trip).
# Pricing: HopBase self-serve price, USD (https://hop-base.com/pricing, also in https://hop-base.com/llms.txt).
# Peak: weekdays 09:00-12:00 and 14:00-18:00 Asia/Shanghai (UTC 01-04, 06-10); off-peak and weekends bill at half.
# Cache reads: this group discounts cached input (off-peak 0.0018); the non-sg ID bills cache reads
# at the official cache price (0.00295 off-peak), which is why the two differ.
# Peak also applies on Chinese public holidays that fall on weekdays: HopBase billing uses the
# weekday windows only, with no holiday exception.
schema = 2
base_model = "deepseek/deepseek-v4.1-flash-1017"

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

[limit]
output = 393_216

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

[cost.text.input]
price = 0.08825
cache = { read = 0.0018 }

[cost.text.output]
price = 0.35295

[[cost.multipliers]]
when = { utc_days = { in = ["mon", "tue", "wed", "thu", "fri"] }, any = [{ all = [{ utc_hours = { gte = "01:00" } }, { utc_hours = { lt = "04:00" } }] }, { all = [{ utc_hours = { gte = "06:00" } }, { utc_hours = { lt = "10:00" } }] }] }
factor = 2
44 changes: 44 additions & 0 deletions providers/hopbase/models/deepseek-v4.1-flash-1017.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,44 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# Limits: 1M context, up to 393,216 output tokens (https://hop-base.com/llms.txt).
# Toggle: thinking.type = enabled|disabled (disabled -> 0 reasoning tokens).
# Effort: reasoning_effort = low|high|max, the native DeepSeek levels; medium is accepted and
# mapped by DeepSeek, none also turns thinking off, an unknown value returns HTTP 400.
# Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Images: a red PNG data URL returned "Red". Structured output: json_schema with strict = true
# returned an enum-valid object. Cache: repeating a ~2.4K-token prefix returned cached_tokens = 2432.
# tool_choice none|required|named all honoured.
# Pricing: HopBase self-serve price, USD (https://hop-base.com/pricing, also in https://hop-base.com/llms.txt).
# Peak: weekdays 09:00-12:00 and 14:00-18:00 Asia/Shanghai (UTC 01-04, 06-10); off-peak and weekends bill at half.
# Cache reads: this ID's group bills cached input at the official cache price, so the off-peak
# cache read is 0.00295 (half of 0.0059); the -sg ID's group discounts cache reads too (0.0018).
# Peak also applies on Chinese public holidays that fall on weekdays: HopBase billing uses the
# weekday windows only, with no holiday exception.
schema = 2
base_model = "deepseek/deepseek-v4.1-flash-1017"

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

[limit]
output = 393_216

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

[cost.text.input]
price = 0.08825
cache = { read = 0.00295 }

[cost.text.output]
price = 0.35295

[[cost.multipliers]]
when = { utc_days = { in = ["mon", "tue", "wed", "thu", "fri"] }, any = [{ all = [{ utc_hours = { gte = "01:00" } }, { utc_hours = { lt = "04:00" } }] }, { all = [{ utc_hours = { gte = "06:00" } }, { utc_hours = { lt = "10:00" } }] }] }
factor = 2
51 changes: 51 additions & 0 deletions providers/hopbase/models/doubao-seed-2-0-mini-260428.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,51 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# Limits: 256K context, up to 128K output (https://hop-base.com/llms.txt); max_tokens=131072 -> 200,
# 131073 -> HTTP 400 "expected a value <= 131072".
# tool_choice none|required|named all honoured.
# Effort: reasoning_effort = none|minimal|low|medium|high|xhigh|max, all accepted live; none and
# minimal return 0 reasoning tokens, so no separate toggle is declared (thinking.type = disabled is
# also accepted). An unknown value returns HTTP 400. Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Images: a red PNG data URL returned "Red"; HopBase documents image input (public URL or base64)
# but not video, so input is limited to text and common image types here.
# Structured output: json_schema with strict = true returned an enum-valid object.
# Pricing: HopBase list price, USD; base rates follow the official Volcano Engine list price.
# Tiers follow HopBase billing, not Volcano's table: a single
# long-context tier when input is greater than 32,768 tokens (long_context.threshold = 32768 in
# https://api.hop-base.com/api/v1/models/pricing); no further tier on this host.
schema = 2
base_model = "bytedance-seed/seed-2.0-mini"

[modalities]
input = ["text", "image/jpeg", "image/png", "image/gif", "image/webp"]
output = ["text"]

[capabilities]
tools = { supported = true, choice = ["auto", "none", "required", "tool"] }
reasoning = { supported = true, effort = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }
structured_output = true
stream_usage = true

[limit]
output = 131_072

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

[cost.text.input]
price = 0.0295
cache = { read = 0.0059 }

[[cost.text.input.overrides]]
when = { context = { gt = 32_768 } }
price = 0.0589
cache = { read = 0.0118 }

[cost.text.output]
price = 0.2942

[[cost.text.output.overrides]]
when = { context = { gt = 32_768 } }
price = 0.5883
35 changes: 35 additions & 0 deletions providers/hopbase/models/doubao-seed-2-1-lite-260915.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,35 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# Limits: 1M context, up to 256K output (https://hop-base.com/llms.txt); max_tokens=262145 -> HTTP 400.
# tool_choice none|required|named all honoured.
# Effort: reasoning_effort = none|minimal|low|medium|high|xhigh|max, all accepted live; none and
# minimal return 0 reasoning tokens, so no separate toggle is declared (thinking.type = disabled is
# also accepted). An unknown value returns HTTP 400. Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Images: a red PNG data URL returned "Red"; HopBase documents image input (public URL or base64)
# but not video, so input is limited to text and common image types here.
# Structured output: json_schema with strict = true returned an enum-valid object.
# Pricing: HopBase list price, USD; follows the official Volcano Engine list price.
schema = 2
base_model = "bytedance-seed/seed-2.1-lite"

[modalities]
input = ["text", "image/jpeg", "image/png", "image/gif", "image/webp"]
output = ["text"]

[capabilities]
tools = { supported = true, choice = ["auto", "none", "required", "tool"] }
reasoning = { supported = true, effort = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }
structured_output = true
stream_usage = true

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

[cost.text.input]
price = 0.1177
cache = { read = 0.0236 }

[cost.text.output]
price = 0.3971
38 changes: 38 additions & 0 deletions providers/hopbase/models/doubao-seed-2-1-pro-260628.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,38 @@
# Verified live 2026-10-31 against POST https://api.hop-base.com/v1/chat/completions with this exact
# model ID: non-streaming, streaming (stream_options.include_usage returns usage), and a
# tool call (tool_choice auto) that returned a well-formed function call.
# Limits: 256K context, up to 128K output tokens (https://hop-base.com/llms.txt).
# tool_choice none|required|named all honoured.
# Effort: reasoning_effort = none|minimal|low|medium|high|xhigh|max, all accepted live; none and
# minimal return 0 reasoning tokens, so no separate toggle is declared (thinking.type = disabled is
# also accepted). An unknown value returns HTTP 400. Relative effort was not benchmarked.
# Reasoning text is returned in message.reasoning_content and delta.reasoning_content.
# Images: a red PNG data URL returned "Red"; HopBase documents image input (public URL or base64)
# but not video, so input is limited to text and common image types here.
# Structured output: json_schema with strict = true returned an enum-valid object.
# Pricing: HopBase list price, USD; follows the official Volcano Engine list price.
schema = 2
base_model = "bytedance-seed/seed-2.1-pro"

[modalities]
input = ["text", "image/jpeg", "image/png", "image/gif", "image/webp"]
output = ["text"]

[capabilities]
tools = { supported = true, choice = ["auto", "none", "required", "tool"] }
reasoning = { supported = true, effort = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }
structured_output = true
stream_usage = true

[limit]
output = 131_072

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

[cost.text.input]
price = 0.8824
cache = { read = 0.1765 }

[cost.text.output]
price = 4.4118
Loading
Loading