Claude Haiku 4.5 vs GPT OSS 20B

GPT OSS 20B is the better all-round choice, scoring 61/100 against Claude Haiku 4.5 (57). It leads on price. Claude Haiku 4.5 wins on capability, inputs & features and context window. The score weighs capability 50%, price 25%, inputs & features 15%, context window 10%.

GPT OSS 20B is cheaper at $0.07 input / $0.25 output per million tokens (median across 37 API providers). Claude Haiku 4.5 costs $1.00 input / $5.00 output per million tokens (official Anthropic API price). At a typical mix of three input tokens to one output token, that is $0.115 per million tokens for GPT OSS 20B versus $2.00 for Claude Haiku 4.5 (17× as much).

Claude Haiku 4.5 scores higher on the Capabilities Index (ECI): Claude Haiku 4.5 142.2 (#102 of 165) and GPT OSS 20B 137.8 (#122 of 165). The confidence ranges of the top two overlap (139.4–143.9 vs 133.1–139.7), so treat the gap as small. On individual benchmarks: GPQA Diamond — Claude Haiku 4.5 71.2%, GPT OSS 20B 60.8%; OTIS Mock AIME 2024–2025 — Claude Haiku 4.5 66.7%, GPT OSS 20B 65.3%.

There are no published SWE-bench Verified results for Claude Haiku 4.5 and GPT OSS 20B yet, so there is no like-for-like coding score. On overall capability, Claude Haiku 4.5 leads, which tends to carry over to coding, but test on your own codebase. Both support tool calling for agent workflows.

Claude Haiku 4.5 has the largest context window at 200,000 tokens, against 131,072 for GPT OSS 20B. Maximum output per response: Claude Haiku 4.5 up to 64,000, GPT OSS 20B up to 32,768 tokens.

Claude Haiku 4.5 accepts text, images and PDFs; GPT OSS 20B accepts text. Claude Haiku 4.5 handles the widest range of inputs.

GPT OSS 20B publishes its weights and can be self-hosted; Claude Haiku 4.5 is proprietary.

Claude Haiku 4.5 is the newest, released Oct 15, 2025. GPT OSS 20B came out Aug 5, 2025. Knowledge cutoff: Claude Haiku 4.5 Feb 28, 2025, GPT OSS 20B Jun 2024.

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Nov 2, 2026.