flagship model · live

Ornith-1.5 35B-A3B

The mid-size mixture-of-experts member of the Ornith-1.5 family. It activates only ~3B parameters per token, yet significantly outperforms its similar-sized peer Qwen 3.6-35B across all coding and agentic benchmarks, and outperforms dense models such as Gemma 4-31B and Muse Glimmer-30B by wide margins on agentic coding.

Ornith AI’s own description of the model, from Ornith-1.5-35B-A3B model card . Served here as ornith-ai/ornith-1.5-35b-a3b.

What Ornith AI built it for

in the vendor’s words

“Ornith-1.5-35B-A3B excels in tool-calling and agentic coding.”
“Ornith-1.5-35B-A3B is optimized for terminal-based coding agents.”

Quoted from Ornith-1.5-35B-A3B model card, retrieved 23 Aug 2026. A reasoning model with no non-thinking arm documented: the card publishes one set of scores, not a thinking/non-thinking split.

Published benchmarks

Ornith AI’s figures, not ours

Ornith-1.5 35B-A3B — scores as published by Ornith AI, retrieved 23 Aug 2026
Benchmark	What it measures	Score	How Ornith AI ran it
Terminal-Bench 2.1 Terminus-2 harness	Agentic terminal coding	67.8	Harbor/Terminus-2 framework, parser=json, temperature=1.0, top_p=1.0, 128K context window; 4-hour timeout, averaged over 5 runs. Mode: reasoning on (the model's only documented mode).
SWE-bench Verified	Real-world software engineering	79	OpenHands harness, temperature=1.0, top_p=0.95, 256K context window; git history removed and network disabled as anti-hacking safeguards. Mode: reasoning on (the model's only documented mode).
SWE-bench Pro	Real-world software engineering, harder split	59.6	OpenHands harness, temperature=1.0, top_p=0.95, 256K context window. Mode: reasoning on (the model's only documented mode).
MCP-Atlas	Tool use over MCP servers	70.2	500-task public subset, 10-minute timeout per task, Claude 4.8 Opus as judge. Mode: thinking mode (the card states all models were evaluated in thinking mode for this row).
ClawEval	Agentic coding on real-user tasks	72.5	temperature=0.6, 256K context window. Mode: reasoning on (the model's only documented mode).
GPQA Diamond	Graduate-level science reasoning	89.2	The card states no separate harness or sampling for this row beyond the family-wide 5-run average. Mode: reasoning on (the model's only documented mode).
HLE no tools	Humanity's Last Exam, frontier reasoning	25.6	Claude 4.6 Opus as the judge model; the card also publishes 33.4 with tools. Mode: reasoning on (the model's only documented mode).

Every score above is Ornith AI’s own published result for this model, copied from Ornith-1.5-35B-A3B model card on 23 Aug 2026 — we did not run these benchmarks and we do not restate them as our own. “All results reported for Ornith-1.5 are averaged over five independent runs.” Each vendor ran its own harness with its own judge, so these tables are not a league table: scores from two different cards are not comparable, even where the benchmark name matches. Each model here shows only its own maker’s table, for that reason.

Recommended usage

what Ornith AI recommends

Sampling

For general tasks, Ornith AI recommends temperature=0.6, top_p=0.95, top_k=20. That is what this endpoint sends when your request omits them, so the default request shape is the vendor’s recommendation rather than ours. An explicit value in your request always wins.

Ornith AI publishes a second set for reproducing the vendor's published benchmark scores: temperature=1.0. Send those yourself if that is your case — they are not the default here.

Thinking

Reasoning is always on in the vendor's own recipes — the card documents no request-side switch to turn it off, and the serving recipes it publishes enable a reasoning parser so the chain of thought comes back in a separate reasoning_content field rather than inline.

On this endpoint: Reasoning is on or off, nothing in between: on by default for hard problems, off with reasoning_effort "none" — the low-latency shape for voice and interactive turns.

Context length

Ornith AI gives the native window as 262,144 tokens, which is what we serve — the full checkpoint window, with input and output sharing it. Beyond the native window the vendor points at YaRN RoPE scaling — factor 4.0 takes the usable window to roughly 1M tokens — but recommends against leaving it on.

Ornith AI’s caution on extending it: “Open-source runtimes implement YaRN statically: the same scaling factor is applied to every request regardless of its length, which can slightly hurt quality on ordinary-length inputs. Only enable rope_scaling when your workload genuinely needs the longer window, and size factor to match it” We serve the native window and do not enable static RoPE scaling, so ordinary-length requests are not paying that cost.

What to point it at

Ornith AI names these workloads for this model:

“understand large codebases, automate tedious work, and ship faster”
“works out of the box with standard agent frameworks”

Recommendations on this page are Ornith AI’s, from Ornith-1.5-35B-A3B model card, retrieved 23 Aug 2026. We default to them rather than to a house preference; where our default would differ, this page says so.

What it does on this endpoint

our measurements · not vendor figures

output speed up to 305 tok/s single stream, measured live

first token 120 ms median, same-region US, reasoning off

context 262K tokens, the native window

These three are ours, measured on the live endpoint — observations, not service guarantees, and not to be confused with the vendor scores above. The protocol, the vantages, the build each figure is pinned to and how to check it against the system_fingerprint in your own response are all on the measurements page. Run your own and use those instead.

Price

per million tokens · no request fee · no subscription

Token rates — Ornith-1.5 35B-A3B
Input	Cached input	Output
$0.25	$0.09	$1.20

One price, no tiers. Cache hits are applied automatically and reported per response as cached_tokens. Full price sheet and worked examples.

Capabilities

ornith-ai/ornith-1.5-35b-a3b

Three API formats, one base URL

OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages — all served from https://api.tiyuvta.ai/v1. Point an existing client at it and set the model id; there is no product-specific SDK.

Tools and structured output

tools and tool_choice, JSON mode, and JSON schema via response_format. The tool-call round trip is part of the release gate on every deploy.

Vision

Images go in as image_url content parts and bill as ordinary input tokens — vision costs nothing extra.

Reasoning: on or off

Reasoning is on or off, nothing in between: on by default for hard problems, off with reasoning_effort "none" — the low-latency shape for voice and interactive turns.

Prompt caching

Continuing a conversation reuses its prefix and bills the cached rate for what actually hit — reported per response, priced per token, no hit-rate promises.

Exactness-gated serving

Speculative decoding is verified byte-identical to plain decode on the serving configuration before a build takes traffic — speed never changes the output. The correctness gates.

Where it fits best

use it for

The default choice: the flagship — fastest first token, highest output speed, lowest price here
Agents and tool-calling loops, where output tokens are the bill
High-throughput pipelines — extraction, generation, summarization at volume
Voice and interactive products: under 200 ms median to the first content token from both US coasts with reasoning off

Need a reasoning-effort dial or video input? See Qwen3.8 27B — same key, same balance.

First request

ornith-ai/ornith-1.5-35b-a3b

shell — default https://api.tiyuvta.ai/v1

curl https://api.tiyuvta.ai/v1/chat/completions \
  -H "Authorization: Bearer $TIYUVTA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "ornith-ai/ornith-1.5-35b-a3b",
       "messages": [{"role": "user", "content": "Hello"}]}'

The default request reasons before it answers, at Ornith AI’s recommended sampling — best quality for hard problems.

shell — low latency reasoning off

curl https://api.tiyuvta.ai/v1/chat/completions \
  -H "Authorization: Bearer $TIYUVTA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "ornith-ai/ornith-1.5-35b-a3b",
       "reasoning_effort": "none",
       "messages": [{"role": "user", "content": "Hello"}]}'

reasoning_effort: "none" makes the first streamed token the answer itself — the measured 120 ms first-token figure above is this request shape.

Create account Agent client configurations Read the docs

Weights and vendor sources: huggingface.co/ornith-ai/Ornith-1.5-35B-A3B, MIT. Ornith AI trained this model; we serve it at its full native context in NVFP4 4-bit weights (self-published GGUF, first NVFP4 of this model). Benchmark scores and usage recommendations on this page come from Ornith-1.5-35B-A3B model card and Ornith-1.5: From Self-Scaffolding to Self-Improvement , retrieved 23 Aug 2026.
Machine-readable metadata: GET https://api.tiyuvta.ai/v1/models for supported parameters, and the provider feed at GET https://api.tiyuvta.ai/models?schema=openrouter.
Rate limits: Per account, for abuse control, raised on request — capacity scales with demand rather than sitting behind a published ceiling.